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

Tillandsia L. Subg. Phytarrhiza (Visiani) Baker (Bromeliaceae) is a distinctive group of about 35 epiphytic species. These exhibit a range of habits from xeric to mesic. The evolutionary relationships of the contrasting adaptations need to be established here as well as in the subfamily as a whole. Relations between the subgenus and other tillandsioids are problematical and phylogenetic reconstruction of its member-species would be facilitated by identification of Phytarrhiza's relative (sister taxon) sharing the same most recent common ancestor with Phytarrhiza. This paper examines the two most likely sister taxa, Subg. Pseudo-Catopsis Baker and Subg. Diaphoranthema (Beer) Baker. Diaphoranthema is rejected as sister taxon. The accepted evolutionary tree, rooted by Pseudo-Catopsis, indicates that most habital evolutionary changes in Phytarrhiza have been between mesic and semi-mesic forms and from mesic to xeric forms. Methods developed for testing specific evolutionary hypotheses are broadly applicable. THERE IS CONSIDERABLE interest in the direction of evolution of species within Phytarrhiza as well as for the entire genus and subfamily because of questions raised by Pittendrigh (1948), Medina (1974), and Benzing and Renfrow (1971 a, b), and discussed by Benzing, Givnish and Bermudes (1985), and Gilmartin (1983), regarding relationships between mesic, tank forms and more strees-adapted "atmospheric bromeliads." Benzing et al. (1985) examined the question: are extreme xeric tillandsioids derived from mesic formsSchimper's (1888) interpretation?; or are mesic, epiphytic tillandsioids derived from xeric precursors-Pittendrigh's (1948) interpretation? Benzing also considered a third interpretation, proposed by Medina (1974) who examined carbon pathways. Medina proposed that both xeric and mesic tillandsioids arose from precursors adapted to conditions of high light intensity and humidity. There is little support for the often held notion that xeric adaptations in general, are more frequently acquired from mesic progenitors than the reverse. Some reasons for the notion are the many examples in the world's deserts ' Received for publication 6 May 1985; revision accepted 4 October 1985. Research supported by NSF Grant BSR 84407573 to the authors. The following individuals provided extremely helpful critical reviews: David Benzing, Phil Cantino, Vicki Funk, Loren Rieseberg, Karen Simmons, Douglas Soltis, and John Utley. Robin Lesher provided technical assistance and plant illustrations were rendered by Sheila Gilmartin. ofxeric specializations of otherwise mesic taxa; for example, coreaceous, linear leafed species of Lycium (Solanaceae). Such xeric adaptations might seem to be difficult to reverse, i.e., unlikely to evolve to a more mesic state. Stebbins (1974) pointed out, however, that any belief that xerophytes, in general, are irreversibly specialized has no foundation in fact. We are left with both courses being essentially equally

ofxeric specializations of otherwise mesic taxa; for example, coreaceous, linear leafed species of Lycium (Solanaceae).Such xeric adaptations might seem to be difficult to reverse, i.e., unlikely to evolve to a more mesic state.Stebbins (1974) pointed out, however, that any belief that xerophytes, in general, are irreversibly specialized has no foundation in fact.We are left with both courses being essentially equally probable.
Geographic areas that might be paricularly prone to harbor evolutionary sequences of a given lineage in both directions, would be regions where pluvial and arid climates have alternated.During the Pleistocene, pulsating, climatic changes occurred at least in some parts of South America (Haffer, 1969;Sarmiento, 1975;Simpson, 1975;Solbrig, 1976;Gilmartin, 1983).
With these ideas in mind, and using phylogenetic reconstruction methods, we examined the possible direction(s) of evolution among members of a closely related group of tillandsioid species.Tillandsia is a large bromeliad genus of mixed habit with its center of distribution in South America.The study group members are all epiphytic or saxicolous, in either case, nutritionally independent from the substrate (Pittendrigh, 1948) (Rumley, 1965;Gilmartin, 1972Gilmartin, , 1973Gilmartin, , 1983)).
We ask the question: within this group, which direction(s) of evolutionary change, if any, occurred most frequently: xerism toward mesic adaptations, or mesic forms toward xerophytes?In this first of a series of papers we seek the answer in taxonomic revisions, and by constructing cladograms and examining patterns of morphological character-state changes.
Phytarrhiza initially appeared to be monophyletic, unlike other Tillandsia subgenera, e.g., Allardtia and Tillandsia, that could not be shown to be natural groups (Gardner, 1982).Tillandsia Subg.Phytarrhiza species all share the uniquely derived petal-character, blades broad and conspicuous (Gilmartin, 1983), and is the only bromeliad group with conspicuous petal-lamina.The three subgenera Diaphoranthema, Pseudo-Catopsis, and Phytarrhiza all have very short styles that are included with the stamens within the corolla.Together these three subgenera constitute a putatively monophyletic group.
The relationship of Phytarrhiza to other tillandsioids has been a difficult question given previously available data and traditional analytical tools.This has not impeded speculation.The tools of modem phylogenetic reconstruction are brought to bear here on the question of relations of Phytarrhiza to other tillandsioids with focus on two of the most likely contenders for sister taxon status.A sister taxon is that evolutionary unit sharing with the group under study the same most recent common ancestor.Its identification is one of the most effective means to determine direction of evolution of characters (Hennig, 1966;Wiley, 1981), i.e., to establish which states are ancentral, which are derived.
Strong tendencies toward parallel evolution are evident within every major group of organisms.Among traits showing parallel evolution, those involving less complexity are more likely to exhibit reversals, i.e., change in more than a single direction, than are those involving greater complexity (Futuyma, 1979) and while both parallels and reversals may occur, often we expect more of the former and fewer of the latter.Stebbins' (1974) concepts of paths of least resistance may help to explain why parallel evolutionary change occurs as frequently as it does among related taxa.
Parallel evolution is very common in flowering plants (Cronquist, 1968;Funk, 1981), and bromeliads are no exception (Benzing, 1980;Gardner, 1982).For example phylogenetically isolated bromeliad taxa have several times independently acquired nectar scales on the ventral petal surface (Smith and Downs, 1974, 1977, 1979).Parallel characters should not be removed however, before constructing trees because they may be useful at branch tips (Funk, 1981).In this work we retain characters regardless of subsequent evidence of parallel changes.By parallel character-state changes, we mean recurrent changes of the same character in the same direction whether it is a recurrent gain or loss.
Certainly, reversals occur frequently.Ownbey and Aase (1955) pointed out following their research with Allium, that some reversals occur especially readily.Loss of a structure is particularly likely to recur when the original evolution of the trait involved several genes, and mutation of any one of these might prevent expression of the trait (Ownbey and Aase, 1955).Relative to the outgroup, these are recurring forward changes or recurring reversals.However, within a closely related group of taxa, such recurring losses would constitute parallels within the group under study.
Three of the most conspicuous, morphological determinates of mesic versus xeric habit (Fig. 1) were not included in the data when forming the groups: leaf-blade width and degree of succulence, and presence and development of a tank.This was by design to avoid circularity, so that resulting cladograms would be based on other data than the very characters for which we hoped to discern polarity, i.e., direction of evolution.(The groups did not change when the analyses included these characters.)MATERIALS AND METHODS-Data were assembled from the Monograph by Smith and Downs (1977), and methods of phylogenetic reconstruction were applied under the principle of maximum parsimony (Wiley, 1981).Maximum parsimony has the goal of reconstructing phylogeny with the fewest possible character-state changes for each character.The directions of change (character-state polarities), were established using the two most likely alternative sister taxa, Subg.Pseudo-Catopsis and Subg.Diaphoranthema.This approach has been described as the outgroup substitution method by Donoghue and Cantino (1984).These authors suggested using multiple outgroups alone and in combination to identify a consensus tree.Our objective is to identify the most likely hypothetical sister taxon, because our central concern is the direction(s) of evolution relative to xerophytes and mesophytes.The study group is analyzed with the two alternative sister taxa that prior research suggested.
Deployment of Diaphoranthema as the sister taxon to Phytarrhiza is implicit in Smith's remarks (1934) regarding the apparent closeness of these two taxa.Characters linking Phytarrhiza and Diaphoranthema include exserted petal blades, presence of floral tube, anthers equalling or exceeding stigma with both sets of organs included within the corolla (Smith and Downs, 1977).However, it is not clear whether Diaphoranthema was thought to be a closely related taxon at the same level as Phytarrhiza, i.e., what would be termed a sister taxon or thought to have evolved from one or several phytarrhizan species, i.e., represent a part of the study group or ingroup in current terms.The latter would mean that Phytarrhiza is not monophyletic, but paraphyletic, and this question is examined.We are often forced to deal with paraphyletic groups during early stages of analyses-and analyses may reveal study groups to be paraphyletic (Platnick and Funk, 1983).
Prior research on the subgenera of Tillandsia and of Vriesea (Gilmartin, 1983)  Every taxon in Pseudo-Catopsis and Diaphoranthema initially was included in all analyses.Subsequently, a single species of P-C, out of the toal of about 48 was excluded.Tillandsia pugiformis L. B. Smith, may have stems, a trait that sets it apart from other species of P-C.Five of 17 species of Diaphoranthema had leaves that were polystichous in contrast to the typical distichous habit.These tend to be polyploids (Till, 1984).The putative sister taxon that was used for the D trees represents only distichous leaved members of this subgenus.
Phylogenetic reconstruction was done with PAUP version 2.0 (Swofford, 1983) on the WSU Amdahl IV.Data were input in standard format.Among the program's available options, the following were used: 1) "closest" determined the sequence in which evolutionary units would be added to the tree; 2) "global branch swapping" identified the shortest trees; 3) "mulpars" searched for multiple equally parsimonious trees using global branch swapping (see above); 4) "apolist" listed the apomorphies, and 5) "chglist" listed changes in each character and the branch(es) where these appeared.Trees were rooted in two ways, the Lundberg (1972) method with "outstates," i.e., states of putative outgroup sister taxon establishing direction of changes, and by including the outgroup in the analyses.
RESULTS-Preliminary to generating trees, 13 phytarrhizan species-groups (alliances) and single species were identified based upon assembled data (Appendix).Several groups appeared to be monophyletic using either P-C or D, or both as the sister group.The groups were selected prior to the computer analyses for phylogenetic reconstruction.Phytarrhiza Group 1, (Phyt I) for example, shares three uniquely derived (apomorphic) states relative to sister taxon P-C.The character-state data are: state 5Bscape bracts none to three, 6B-leaves distichous, 1 2A-leaf-blades terete (Fig. 2).Some putative evolutionary units (EUs) are set apart by combinations of traits, e.g., T. humilis Pres together with group VI has the unique set of character-states: 1B, 2B, 3A, 7B, i.e., floral bracts, elliptic and ecarinate, stems long and conspicuous, and inflorescence compound.Employing rooting by P-C, character-states 1A and 3B would represent derived states; relative to D, character-state 7A has the derived state while for both putative sister taxa, 2B is ancestral.Table 1 lists all 15 characters and their states.Figure 2 shows the character-state values of all putative evolutionary units including the two putative sister taxa to Phytarrhiza (P-C, Diaph 2).
Monophylesis of Diaphoranthema together with three species of Phytarrhiza is supported by the derived character-states, distichous leaves and few scape-bracts.The three phytarrhizan species that share these states are T. bandensis Baker, T. mallemontii Glaziou ex Mez, and T. crocata (E.Morren) Baker.
Our hypotheses concerning which direction of change occurs more readily are indicated in Table 1 by an asterisk adjacent to the state from which a change is presumed to most easily originate.We hypothesize, for example, that an evolutionary change in character five from A, scape-bracts several or many to B, none to three, will occur more frequently than change from no or few scape-bracts to several scapebracts.This is not to say that we are presuming polarities for our study group.On the contrary, we will test our hypotheses against the objectively generated trees and the polarities of characters will be determined by the effects of outgroups on the ensuing trees.
It was only after alliances were identified and analyses were run that each alliance was classified as mesic or xeric.As it turned out, using leaf-blade width as the indicator, each group was uniform in this regard although the character was not part of the matrix (Table 1), nor did we consider it when drawing up the groups.These groups depended solely on the selected characters in Table 1.
Character eight (foliar scales spreading or adpressed) may have a bearing on mesic versus xeric habits and habitats.It is well documented, e.g., that the spreading foliar scales of xeric tillandsioids are ideally suited for water absorption (Tomlinson, 1969;Benzing, 1980).Foliar scales are present in all tillandsioids from xeric sites, but species from mesic habitats may also have a dense covering of scales, e.g., T. tetrantha var.aurantiaca (Griseb.)L. B. Smith (Gilmartin, 1972), T. magnusiana Wittm., and T. argentea Griseb.(Gardner, 1982).Dense foliar scales and narrow, thick leaves of T. bulbosa Hooker give this species the appearance of a strong-xerophyte; yet, apparently it grows equally well in shady, moist habitats (Benzing, 1980).There is very imperfect concordance of P C 1 Habit types, mesic (m), semi-mesic (s), and xeric (x) could be determined for the terminal groups on trees, but at internal nodes more than one possibility exists.Possible habits have been entered at the nodes in Fig. 3 and 4.

AUREA B B B A A A A A B B A B B B B PHYT 7 B B B A A A B B A B B B B B ? PHYT 8 B B B A A A B A A B A B A B ? HUMULIS B B A A A A B A A B B B B A ? CAERULEA B B A A
In Fig. 3a   acter-state changes, only the direction of change.
Trees using the 2 sister taxa were equal in length, 24 character-state changes.Therefore, ways were sought to evaluate trees of equal length.
A new method was employed to gauge trees by their stability, i.e., number of equal lengthened trees and by resolution of evolutionary units.A set of fewer, equally parsimonious trees that are more fully resolved is preferred because they are more precise and more falsifiable (i.e., testable with additional data) than are a larger number of equally short trees that are less resolved.This is an extension of the parsimony principle (Occam's Razor) which cladistics applies usually to the number of steps of character-state changes in trees.
Total parallelism, i.e., convergence and reversals, homoplasy, is indicated by the CI, consistency index, (number of parallels and reversals divided by number of character-state changes), and PAUP treats equally parallels and reversals.For instance, a tree produced with 10 -characters and 20 character-state changes has a CI of 0.50.Some changes must be parallels or reversals in this case, but the relative contributions to homoplasy of paral- Trees maximizing parsimony (24 characterstate changes) had the same consistency index of 0.58.Figures 3a and 4a are the two most frequent, shortest trees for 1 1 evolutionary units of Subg.Phytarrhiza.These are rooted in Fig. 3 with outstates of Pseudo-Catopsis (P-C) and in Fig. 4 with Diaphoranthema (D).The broken line in each case indicates the concensus, i.e., all EUs within the broken line uniformly appeared exactly as shown in every equally short tree.The three classes of habits are indicated: m (mesic), s (semi-mesic), x (xeric).
All variants of the two basic configurations are shown in Fig. 3 (a-d), and Fig. 4 (a-g).These variants involve the branch supporting groups II, III, IV, V and VII in Fig. 4 and only II, III, IV and V in Fig. 3.The concensus tree for the cladograms generated with P-C as the outgroup (Fig. 3) includes seven phytarrhizan EUs: I, VI, VII, VIII, T. humilis, T. aurea and T. caerulea, plus Diaphoranthema.The exact sa 'me configuration of these evolutionary u-nits ap-peared in all four trees rooted with P-C outstates or rooted by P-C when it was included in the study group.
There were seven trees that were rooted by Diaphoranthema or by its outstates.This is three more than the set of P-C trees.Their consensus incorporates six phytarrhizan units, I, VI, VIII, T. humilis, T. aurea and T. caerulea and Diaphoranthema (broken line Fig. 4).The same configurations occurred when P-C and D were used together as an outgroup, though such rooting could not produce a tree because the ingroup was not monophyletic when both of these were used as the outgroup.
Equally parsimonious trees (phylogenetic inferences) may differ in at least four ways.P-C and D trees differed in: 1) total number of trees, 2) number of parallels and total reversals, 3) the number of resolved groups, and 4) the particular characters that underwent changes (Table 2).In fegard to 1), there were four possible P-C trees and seven possible D-trees.Regarding 2), each of three possible types of character-  2a identifies the number of characterstate changes for the two principle D and P-C trees.Total numbers of each type (single, parallel and reversal) were 13, 7, 4 on the P-C tree, and 14, 4, 6 on the D tree.There were two more reversals and three more parallels on the D than the P-C trees.The ratio of parallels to reversals on the P-C tree was 7:4 = 1.75; on the D tree it was 4:6 = 0.66.Regarding 3), the number of resolved groups of Phytarrhiza on D trees, 5-7 groups are supported by one or more uniquely derived character-states (Fig. 4).On P-C trees, 6-7 phytarrhizan unit groups are supported (Fig. 3).Conversely, 4-6 and 4-5 groups on D and P-C trees respectively, were unsupported.Three nodes are un-resolved in the consensus trees for both sets of trees (Fig. 3, 4).
Regarding our hypotheses about frequencies of character-state changes, three of the changes in character 3 (stem elongation) on the P-C tree and two on the D tree supported our hypothesis about frequency of character-state change (Table 1, 2b).The D tree had a single change in the opposite direction as well.Character 4 on the D tree agreed with our hypothesis, but disagreed on the P-C tree.Our hypothesis about character 7, inflorescence simple or compound was not supported by either tree.Characters 5, 6, 9, and 13 on the P-C tree consistently showed change in concordance with our hypotheses.These hypotheses must be rejected by the changes on the D tree.A principal reason for stable cladograms is good data.When data are inaccurate as to polarities or there is little support at nodes or there is much homoplasy, cladograms tend to be unstable, meaning they change greatly with small changes in the data (Coombs et al., 1981).It follows that sets of equally parsimonious trees may be evaluated by differences in relative stability (consistency).Those with more stability must be preferred as being more likely to be based on good data.
Of two or more equally parsimonious cladograms, i.e., having the same number of character-state changes, we prefer the one(s) apparently based on better data.We use degree of stability, meaning numbers of different, equal length trees and number of resolved nodes as the pointer to help identify trees that are based on the best data.The characters and characterstates are the same but a polarity change is considered to result in different data.Thus, one polarity may produce better data than another polarity even though the character-states remain the same.
Of the two sets of trees generated with the two alternative polarities, trees using P-C to polarize characters were more fully resolved.There were four P-C trees but seven D-trees.More groups on the P-C trees were resolved than the D trees.The fewer trees and slightly improved resolution of the P-C trees implies greater simplicity and testability of these reconstructions.Furthermore, results suggest a need to examine I, IV, VIII, and T. caerulea, that were unresolved in nearly every tree regardless of which of the two hypothesized sister taxa were used to polarize characters.It remains to be seen if any other characters would support monophylesis of the problematical groups.
On the shortest trees to these eleven alliances, out of 24 changes, the D tree included 6 reversals, among 6 characters and 4 parallels among 3 characters (Table 2b).The P-C tree had four reversals among three characters and seven parallels among five characters.We might ask how our numbers of parallels and reversals compare with a hypothetically best tree of the same length.
We hypothesized before the analysis that the best tree for these taxa is one having the fewest possible parallels, here this would be 10 among the 24 changes and no reversals (Table 2a) in the 14 characters.The differences between P-C and D trees and the hypothetical tree are 8 and 12 changes, respectively.The P-C tree which we have selected on other grounds supports our hypothesis of minimal reversal.
The ratio of number of parallels to reversals for four problematical characters (3, 7, 13, 14) that showed homoplasy and differed in numbers of parallels and reversals, is 6:2 = 3.0 for the P-C trees and 3:4 = 0.75 for the D trees.Our prior hypothesis about the ratio of parallels to reversals is more in concordance with the P-C tree than the D tree, rejected on other grounds.
Character 4, relative expansion of lamina of petal blades, showed a single change on the P-C tree from A to B at the node subtending Diaphoranthema and a single change from B to A at the base of the D tree.As a result of these analyses, we came to realize that the petals of Diaphoranthema are much more like those of Phytarrhiza than of Pseudo-Catopsis, though not nearly so conspicuous as in Phytarrhiza.Petals of Diaphoranthema species typically become somewhat narrow below the blade, whereas petals of Pseudo-Catopsis are ligulate without any apparent constriction or readily identifiable blade.The cladistic analyses helped to focus our attention on the similarity in petals of the two subgenera, Phytarrhiza and Diaphoranthema.
Diaphoranthema and the study group-Trees rooted by outstates of P-C that did not include Diaphoranthema were shorter but had more homoplasy than trees that did include Diaphoranthema.Homoplasy in characters 2, 8, 1 1 and 1 5 occurred when Diaphoranthema was not included but not when Diaphoranthema was included.Thus, although the length of the trees was 22 when Diaphoranthema was omitted (shorter than the 24-step trees including this subgenus) homoplasy was reduced (higher consistency index) when Diaphoranthema was included.Overall homoplasy as given by the consistency index was 0.58 with Diaphoranthema and 0.54 when Diaphoranthema was omitted.
We conclude as a result of these analyses that Phytarrhiza sensu Smith and Downs is paraphyletic.Diaphoranthema is monophyletic.W. Till (pers.comm.)found consistency of the stigma types of Brown and Gilmartin (1984) in species of Diaphoranthema, further supporting Diaphoranthema as a monophyletic group.
Questionable taxa -The taxa Tillandsia cacticola, T. purpurea, and T. duratii were included initially.Although these three taxa are considered within Phytarrhiza, they were not included in the final analysis.Tillandsia duratii is particularly problematical as its inclusion in initial analyses resulted in greatly increased variation in topology regardless of polarities.Tillandsia duratii is a floater, appearing in a number of different positions on various trees.Such variable placement is consistent with a hybrid origin and perhaps this taxon, a particularly large, robust member, is a polyploid hybrid.This is being checked currently.It was omitted from these analyses whose principal goal was to identify the sister taxon to the study group.
In most of the P-C trees maximizing parsimony, T. cacticola and T. purpurea directly subtended the consensus branch.Inclusion of this pair awaits more information about T. purpurea's putative relative, T. straminea HBK.Tillandsia straminea initially was placed by Smith in Subg.Allardtia, moved by Gilmartin (1972) to Phytarrhiza, and subsequently reduced to synonomy with T. purpurea in Phytarrhiza by Smith and Downs (1977).
The sister taxon-Among traits of the accepted sister taxon, Pseudo-Catopsis, are the following: floral bracts ovate to triangular in shape, ecarinate; plant with no apparent stem at maturity; scape and scape-bracts well developed; leaves polystichous; inflorescence compound, plants without roots at maturity; petals ligulate with blades not at all or scarcely distinct; petal blades inconspicuous and scarcely or not at all exserted beyond the calyx; pistil mostly exceeding anthers.Changes relative to habit on the rejected D tree would also lend some credence to the notion that the mesic habit changed in Phytarrhiza more frequently toward xerophytism than the reverse.While the inferred ancestor for the rejected D tree is xeric; within the study group, several changes are from mesic to semi-mesic, and there is but a single possible change from xeric to mesic.

CONCLUSIONS-Relationship between xerophytes and mesophytes within
We have been able to reject the hypothesis of Subg.Diaphoranthema as sister taxon to Tillandsia Subg.Phytarrhiza in spite of its readily apparent close relationship with the latter.Indeed, these analyses support the inclusion of Subg.Diaphoranthema within a monophyletic group consisting of alliances of species assigned to Phytarrhiza.This seems to call for an eventual change in subgeneric circumscriptions.Ultimately, with additional data on floral architecture, particularly that ofthe gynoecium and androecium, chromosome cytology, and stable carbon isotopes, phylogenies will be inferred and with some modifications, these groups may be recognized as sections.Research to obtain required data is in progress (e.g., Brown and Gilmartin, 1984;Brown et al., 1984).
The 11 species groups are working-groups and more data is required before these should be formalized as Sections.In every P-C tree (Fig. 3 Fig. 1. a) Typical tillandsioid of mesic habit; b) semi-mesic habit; c) xeric habit.

[
implicated Subg.Pseudo-Catopsis as an alternative sister taxon to Phytarrhiza.Characters linking all

Fig. 2 .
Fig. 2. Data matrix (characters are columns, rows are putative evolutionary units) for 13 phytarrhizan groups and putative sister taxa.A question mark designates an unscorable character.Trees using Diaphoranthema as the outgroup used Diaph 2.
, the inferred ancestor for the entire group could only be mesic (m) or semi-mesic (s) as indicated.Change from xeric (x) to mesic could have occurred once on the P-C tree at the node subtending unit VII.The alternative, no change from x, is also possible at this node.In addition, several possible changes between mesic and semi-mesic are evident on the four possible P-C trees.On the seven D trees (Fig. 4a-g), the inferred ancestor is xeric.At the node subtending T. humilis there is change from x to m.Two possible changes from an internal node with m to s at a terminal group are evident on the D trees.Phylogenetic reconstructions with PAUP produce undirected trees, that is, positions of the root do not alter the total number of char- union: * A = free; B = connate at least in part.10.Sepal symmetry: A = asymmetric; * B = symmetric.11.Sepal length: A = mostly not over 12 mm; B = 12 mm or longer.12. Leaf-blades: A = terete; B = ligulate to triangular,

Fig. 3 .
Fig. 3. Phylogenetic reconstructions for 11 alliances of Tillandsia Subg.Phytarrhiza, plus Diaphoranthema rooted with Pseudo-Catopsis (P-C trees).The number of character-state changes is proportional to branch lengths (0 to 5): a) entire tree with consensus circumscribed by a broken line; b) through d) represent all of the variants.Synapomorphies are indicated in a).Unresolved nodes are indicated by two dots.
DISCUSSION-Two sets oftrees of equal length (24 character-state changes, consistency index 0.58) were generated.One set polarized char-[Vol.73 is the sister taxon, not Diaphoranthema.Progenitors are mesic or semi- mesic and derivatives tend to be xeric as shown on P-C trees.The xeric members of the concensus group are derived from a mesic or xeric ancestor shared with group VII on the accepted P-C trees.We conclude that xerophytism with-have occurred in both directions.But we have no evidence of change from semi-mesic to xeric adaptations.Thus, in these taxa the semi-mesic adaptation does not appear to be intermediate between the plesiomorphic, mesic and apomorphic, xeric adaptations.
) alliances III and IV share a mesic or semi-mesic common ancestor with a mesic alliance, either V or VII, and usually the three alliances, III, IV, and V do so.Uniformly, the xeric concensus group (I, VI, VII, VIII, T. aurea, T. caerulea, T. humilis) has a xeric or mesic most recent common ancestor with the mesic alliance VII.Our hypothesis regarding expected relative frequencies of directions of change for character 3, presence or absence of stems, was supported by every cladogram including those with the two alternate roots.Change from compact to elongate stems, the direction that we hypothesized, occurred five out of six times (Table 2b).Hypotheses concerning other characters were not sustained by both sets of cladograms.However, hypothesized directions for characters 5, 6, 9, 13 were sustained by the P-C tree.The P-C tree did not sustain our hypothesis that character 4 changes more readily from B (indistinct petal blades) to A (distinct, and conspicuous blades) than the reverse.We therefore reject this hypothesis and accept an alternative hypothesis.This research has suggested to us that character 4 should consist of three rather than two states, one state being ligulate without any distinctive petal limb, a second state being a recognizable but small limb, and the third state, very distinct and conspicuous petal limbs.Results with character 7 (inflorescence compound or simple) showed nearly equal frequency in either direction and we reject our hypothesis concerning this character.Establishing correct character polarities of alliances of species, of Subg.Phytarrhiza is extending our understanding of evolution of the mesic and xeric habits in bromeliads, an active area of investigation (e.g., Benzing et al., 1985).Once a species is channeled into one or the other of the habit-types, further speciation tends to remain within the confines of this channel.This seems to be true, in particular for the xeric end of the spectrum within Phytarrhiza.However, speciation may involve evolutionary jumps from the mesic evolutionary channel to less mesic or to xeric modes and apparently this has occurred at least twice in Phytarrhiza, from mesic ancestors of the alliances V and VII.The utility of these applications of cladistic methodology under assumptions of maximum parsimony, lies in their power to distill patterns of similarities and differences down to a few hypothesized reconstructions, principally two in this research.It enabled rejection of one of these in favor of the other by relative degrees of resolution and by consistency (the relative numbers of equally parsimious trees).Phylogenetic analysis can incorporate a desirable degree of objectivity, and can help test hypotheses about specific characters.With the efforts of the early cladists and those who have been developing the methods up to the present, the potential power of the method is significant and trichomes in Brocchinia reducta (Bromeliaceae) and their evolutionary and systematic significance.Syst.Bot.10: 81-91., AND A. RENFROW.197 1a.The significance of photosynthetic efficiency to habitat preference and phylogeny among tillandsioid bromeliads.Bot.Gaz.132: 19-30.