Ant-plant mutualisms are abundant in the tropics and are popular models for ecological study, but investigating the origin and evolution of such systems requires a phylogenetic framework. A common ant-plant mutualism in the Neotropics involves the genus Cecropia, a group of fast-growing pioneer trees that are important in forest regeneration. Relationships between genera in the tribe Cecropieae (Urticaceae), including Cecropia, Coussapoa, Musanga, Myrianthus, and Pourouma, are unknown and are necessary to investigate the evolutionary history of the Cecropia-ant mutualism. Bayesian phylogenetic analyses of the NADH dehydrogenase (ndhF) chloroplast gene region, the 26S region of nuclear ribosomal DNA, and an exon-primed intron-crossing DNA region support the position of non-myrmecophytic African Musanga within a paraphyletic Cecropia. Neotropical Pourouma and Coussapoa are supported as sister taxa with African Myrianthus as their closest relative. Although it remains uncertain whether myrmecophytism was the ancestral condition of the Cecropia clade, a close relationship between non-myrmecophytic Cecropia sciadophylla and Musanga suggests that the loss of ant associations did not accompany African colonization.
Ant plants (myrmecophytes) are found in many terrestrial ecosystems and are especially abundant in the tropics. They have served as models for the study of mutualism in general (Trager et al. 2010), contributing significantly to current understanding of the nature of reciprocal benefits, how benefits vary among partners, and what costs are associated with partnerships among species (Bronstein 1998). Ecological studies of ant plants have suggested that fitness advantages and the specificity of interactions might have played roles in plant diversification (Davidson and McKey 1993; Lengyel et al. 2009; Weber and Agrawal 2014). However, inferred histories of myrmecophytism are often limited by the extent of systematic knowledge. Phylogenetic information is needed to identify the origins of myrmecophytism and to infer how ant-plant mutualisms might have changed over evolutionary time. If hypotheses about ancient species interactions and diversification rates are to be tested, it is especially important to resolve phylogenetic relationships among ant plants and related non-myrmecophytes at taxonomic levels above the species (Weiblen and Treiber 2015).
A common ant-plant mutualism in the Neotropics involves trees of Cecropia Loefl. and ants of the genus Azteca Forel. Cecropia are fast growing pioneers that play an important role in tropical forest regeneration after disturbance through-out Central and South America. Cecropia provide nesting sites (domatia) and nutritious Müllerian bodies for their ant inhabitants (Berg and Franco-Rosselli 2005; Dejean et al. 2012). Host trees may benefit from the presence of aggressive ants that prey on insect herbivores, prune competing vegetation, and deposit nitrogen-rich debris (Sagers et al. 2000; Bronstein et al. 2006). Ecological studies have speculated about the origin and adaptive significance of this mutualism (Latteman et al. 2014) and, although Cecropia was revised by Berg and Franco-Rosselli (2005), a sister group has yet to be identified. Cecropia is a member of the tribe Cecropieae (Urticaceae), which also includes Coussapoa Aubl., Musanga R. Br., Myrianthus P. Beauv., Poikilospermum Zipp. ex Miq., and Pourouma Aubl. Recent molecular phylogenetic studies have shown that Cecropieae is not monophyletic and that Poikilospermum should be transferred to Urticaceae Juss. (Monro 2006; Hadiah et al. 2008; Wu et al. 2013). Our study was motivated by the need to clarify phylogenetic relationships among the remaining five genera as a basis for investigating the origin of myrmecophily in Cecropieae.
The tribe has a long and complex taxonomic history owing to patterns of morphological intermediacy between Moraceae Gaudich. and Urticaceae. The first reference to Cecropieae is attributed to Dumortier (1829), who included Cecropia and Coussapoa within the family Artocarpideae Dumort. Gaudichaud (1830) included Cecropieae in his “Urticées vraies” (true Urticaceae) based on the orthropous orientation of the ovule, along with the tribes Boehmerieae Gaudich., Elatostemateae Gaudich., Forsskaoleeae Gaudich., Parietarieae Gaudich., and Urereae Gaudich. Pourouma was assigned to a monotypic Pouroumeae Gaudich. in a different group of “Urticées,” whereas Trécul (1847) included Pourouma with Cecropia and Coussapoa in the family Artocarpeae. Engler (1889) placed Conocephaleae Trécul including Cecropia, Coussapoa, Pourouma, Myrianthus, and Poikilospermum in the subfamily Conocephaloideae (Moraceae). Corner (1962) suggested the transfer of Conocephaloideae from Moraceae to Urticaceae, based on the basal placentation of the orthropous ovules. However, Berg (1978) proposed a new family Cecropiaceae on the grounds that straight stamens in bud and basal, sub-basal, or (sub)orthropous ovules were diagnostic. Romaniuc-Neto (1999) proposed that all genera of Cecropieaceae sensu Berg (1978) be regarded as a subfamily of Moraceae except for Poikilospermum which he placed in the Urticaceae. Molecular phylogenetic studies supported the placement of Cecropieae in the Urticaceae (Datwyler and Weiblen 2004; Zerega et al. 2005; Monro 2006; Hadiah et al. 2008; Clement and Weiblen 2009; Wu et al. 2013) and the most recent review of Urticaceae nomenclature validated Cecropieae Gaudich. as the name for the tribe (Conn and Hadiah 2009).
Disagreement among taxonomists over the phylogenetic position of Cecropieae can be attributed to conflicting patterns of morphological similarity in Moraceae and Urticaceae. The tribe includes dioecious trees, shrubs, and hemiepiphytes with aerial or stilt roots, a reduced system of clear latex-bearing canals, spiral phyllotaxis, amplexicaul stipules, terminal inflorescences in either cymes, fascicles, spikes, or globose heads and staminate flowers with straight filaments (Table 1, Fig. 1). None of the recent molecular studies included all genera of Cecropieae in a comprehensive analysis of the tribe nor has morphology of the tribe been examined in a phylogenetic framework.
Cecropia, Coussapoa, and Pourouma are distributed in the Neotropics with approximately 80% of the species occurring in the Amazon region. Musanga and Myrianthus are Afrotropical with most species in the rainforests of the western coast. Cecropia and Musanga are similar morphologically and ecologically, but Musanga lacks the domatia and Müllerian bodies that are common among the Cecropia species engaged in mutualism with ants. Ants are associated with most but not all Cecropia species, whereas only a few species of Coussapoa (O'Dowd 1982; Berg et al. 1990) and Pourouma (Bonsen 1990) are reported to be myrmecophytes. Myrianthus has morphological traits similar to Pourouma, but ants are entirely absent from Myrianthus and Musanga.
Understanding the history of myrmecophytism in Cecropieae requires knowledge of relationships among five genera that until now have not been included in a comprehensive phylogenetic analysis. We estimated a Cecropieae phylogeny using both nuclear and chloroplast DNA regions and compared our findings to a phylogeny based on morphology. Our phylogenetic approach provides a framework for more thoroughly investigating the biogeographic history of myrmecophytism in this tribe.
Materials and Methods
Taxon Sampling—We examined 24 samples representing the five genera included in the core Cecropieae (Cecropia, Coussapoa, Musanga, Myrianthus, and Pourouma) and five outgroup taxa representing four of the five other Urticaceae tribes (Boehmerieae, Elatostemateae, Parietarieae, and Urticeae Lam. & DC) (Appendix 1). Among ingroup taxa, the sampling intensity was scaled with the size of the genus (i.e. the greatest number of samples was analyzed for Cecropia, the largest genus in the tribe).
Classification, species richness, distribution and distinguishing features of Cecropieae (Urticaceae).
Sequencing—Three regions were sequenced, including the ndhF cpDNA region (Zerega et al. 2005), the 26S nuclear rDNA region (Olmstead and Sweere 1994), and the nuclear exon-primed intron-crossing (EPIC) marker A16170b, which was developed for Moraceae (Yao et al. 2013). These regions were chosen to include a slowly evolving locus (26S) for resolving phylogentic relationships at higher taxonomic levels, as well as variable and more rapidly evolving loci (ndhF and EPIC), which are more useful at lower levels. DNA was extracted using the Qiagen DNeasy plant mini kit (Valencia, California) with 20 mg of silica gel preserved leaf fragments or herbarium specimens. The PCR amplification of the ndhF chloroplast region was performed in two separate reactions using primer combinations ndhF8f—ndhF1318r and ndhF972f-ndhF2110r (Olmstead and Sweere 1994). Amplification conditions followed those outlined in Zerega et al. (2005) and thermal cycling conditions were 94°C for 1 min followed by 35 cycles of 95°C for 30 s, 46°C for 1 min, and 68°C for 1 min 30 s with a final extension of 72°C for 7 min. The 26S region was amplified in a single fragment using forward and reverse primers previously developed for Moraceae (Zerega et al. 2005). The PCR amplification of the EPIC marker was performed using primers and conditions outlined in Yao et al. (2013). PCR products for all regions were cleaned by ethanol precipitation and quantified using a NanoDrop 2000c (Thermo Scientific Inc., Waltham, Massachusetts). Sequencing was performed in 10 µL reactions using Big Dye sequencing reagents and protocols (Applied Biosystems, Foster City California), and data were collected using an ABI Prism 3730xl DNA analyzer (Applied Biosystems). Previously sequenced samples from Zerega et al. (2005) were obtained from Genbank for ndhF (AY289253, AY289254, AY289256, AY289257, AY289259–AY289264, and AY289266) and 26S (AY686767–AY686772, AY6868774, AY6868776, AY686780, AY6868782, and AY6868835). Sequences were edited and aligned in Geneious v6.1.7 (Kearse et al. 2012), with manual adjustments in Se-Al v2.0a11 ( http://tree.bio.ed.ac.uk/software/seal/) when necessary.
Morphology—We examined morphological features that varied at or below the tribal level and have been used to distinguish groups in the literature (Table 1, Supplemental Appendix S1). Hadiah and Conn (2009) showed that morphology of Urticaceae provides phylogenetic information in some groups. We scored a matrix of 40 discrete characters for the taxa listed in Appendix 1. We collected data from both field observations (A. L. Gaglioti) and the systematic literature. Information on vegetative morphology was gathered from Guérin (1923), Metcalfe and Chalk (1950), Hickey (1973), Radford et al. (1974), Sorsa and Huttunen (1975), Barth (1976), Bonsen and Welle (1983), Barth (1984), Humphries and Blackmore (1989), Bonsen (1990), Welle et al. (1992), Romaniuc-Neto (1999), Clement (2008), and Clement and Weiblen (2009). Anatomical literature informed our scoring of glandular trichome characters (Metcalfe and Chalk 1950; Gangadhera and Inamdar 1977; Kachroo and Bhat 1981; Setochi et al. 1993). Reproductive characters and states were based on taxonomic literature (Gaudichaud 1830; Font Quer 1985: multiple characters; Chew 1963; Ruiter 1976; Berg 1978; Berg et al. 1990; Chen et al. 2003; Berg and Franco-Rosselli 2005). Taxa were scored at the species level and if the morphological state was either unknown, inconclusive based on the literature, or inapplicable, the character was coded as missing.
Phylogenetic Analysis—Bayesian phylogenetic analysis of morphology was conducted using Mr. Bayes v.3.2.2 (Huelsenbeck and Ronquist 2001; Ronquist et al. 2012) on the CIPRES Science Gateway (Miller et al. 2010) with a Markov k model and a gamma distribution. Four (one cold and three hot) Markov chain Monte Carlo (MCMC) simulations, swapping at default settings, were run for five million generations while sampling every 500 generations until the average deviation of split frequencies fell below 0.01. The posterior distribution of trees was summarized by >50% majority rule consensus tree after discarding the first 25% of the sample as burn-in.
Phylogenetic analyses of DNA sequences were also performed using Mr. Bayes on the CIPRES Science Gateway (Miller et al. 2010), but with a GTR substitution model, gamma-distributed rate variation across sites, and a parameter for the proportion of invariable sites. Four (one cold and three hot) MCMC chains were run for five million generations while sampling every 500 generations and until the average deviation of split frequencies fell below 0.01. Once more, the posterior distribution of trees was summarized by >50% majority rule consensus tree after discarding the first 25% of the sample as burn-in. Analyses examined each DNA region alone, (1) ndhF, (2) 26S, and (3) EPIC, plus (4) a concatenated dataset including all three regions and all taxa listed in Appendix 1, and (5) a concatenated dataset of all three regions but excluding taxa for which any of the three DNA regions were unavailable.
Ancestral State Reconstruction—Ancestral state reconstruction of myrmecophytism was performed in Mesquite v. 3.02 (Maddison and Maddison 2015) on the posterior distribution of trees from the concatenated dataset including all three regions and all taxa. Trees were drawn from the posterior distribution and ancestral states were estimated onto each of 400 trees using maximum likelihood with a Mk1 model of character evolution. Uncertainty associated with unresolved nodes in the majority rule consensus tree was examined by filtering trees in Mesquite to separate trees with C. sciadophylla and Musanga sister to the rest of Cecropia from the posterior trees with trees where C. sciadophylla and Musanga were arranged otherwise. Ancestral states were estimated as above and the probability of a myrmecophytic ancestor of the Cecropia clade was recorded for a random sample of 200 trees for each posterior tree type.
Morphological Phylogeny—The Bayesian consensus tree largely supported prior taxonomic groupings (Fig. 2, Appendix 1). Three genera of Cecropieae were supported as monophyletic. A highly supported clade included Musanga and monophyletic Cecropia. There was less support for a clade including Myrianthus and monophyletic Pourouma. Coussapoa had low clade support and appeared to be sister to the rest of the core of Cecropieae, while Poikilospermum was positioned between the outgroup and the ingroup. The equivocal position of Poikilospermum is consistent with conflicting morphology, in which the dioecious breeding system and hemi-epiphytic habit are reminiscent of Cecropieae, while other characteristics such as unlignified vessel elements, dimorphic wood fibers, stipules not fully amplexicaul, and stamens with inflexed filaments are strikingly similar to Urticeae.
Molecular Phylogeny—The aligned ndhF dataset including all 29 samples was 2,046 base pairs (bp) in length, contained 182 variable positions, and 69 phylogenetically informative positions among the ingroup taxa (3%). For three of four Myrianthus samples (Mwangoka 3151, Birnbaum 913, and Birnbaum 917), complete sequences could not be obtained from the herbarium material and sequences were considerably shorter (325–731 bp). The 26S alignment of all 29 samples was 1,001 bp in length with 49 variable and 30 phylogenetically informative characters among the ingroup (3%). The EPIC dataset was smaller than the other two because sequences could not be obtained from 13 of 29 samples despite repeated attempts at PCR optimization. The EPIC alignment of 371 bp had 46 variable and 17 phylogenetically informative characters among the ingroup (4.6%).
Bayesian results for ndhF and 26S strongly supported the exclusion of Poikilospermum from Cecropieae despite the DNA regions yielding different branching order among Poikilospermum and the outgroups Laportea (Urticeae) and Pilea (Elatostemateae). Limited outgroup sampling and sub-stantial sequence divergence among these taxa are likely to account for branching order among these three taxa being the only highly supported difference between the ndhF and 26S topologies (Soltis and Soltis 2004; Bergsten 2005). Bayesian consensus trees from ndhF and 26S each supported the monophyly of Coussapoa, Myrianthus, and Pourouma, whereas Cecropia was not monophyletic due to the embedded position of Musanga (Figs. 3, 4). The EPIC phylogeny was otherwise poorly resolved and yielded no highly supported conflicts with the ndhF or 26S phylogenies apart from a long branch uniting divergent sequences from Coussapoa nymphaeifolia and Pourouma tomentosa.
The combined analysis of the three DNA regions strongly supported the monophyly of each Cecropieae genus except for Cecropia. There was also strong support for a Cecropia/ Musanga clade (hereafter Cecropia sensu lato) and the position of non-myrmecophytic Cecropia sciadophylla as sister to Musanga (Fig. 5). The Bayesian consensus further suggested that C. sciadophylla plus Musanga might be sister to the rest of Cecropia. Sister to Cecropia s. l. was a clade comprising African Myrianthus and Neotropical Coussapoa plus Pourouma. Simultaneous analysis of a subset of 16 taxa with complete sequences for all three DNA regions also strongly supported the monophyly of Cecropia s. l. and the position of C. sciadophylla and Musanga sister to the rest of the clade. There was low support for Myrianthus as the genus sister to the Cecropia/Musanga clade. Finally, Coussapoa and Pourouma were highly supported as sister to the rest of the tribe.
Origin and Loss of Myrmecophytism—Maximum likelihood estimates for the ancestral condition of Cecropia s. l. varied depending on which trees were sampled from the Bayesian posterior distribution. The probability of a myrmecophytic common ancestor for the group ranged from 0.01–0.99 according to the position of non-myrmecophytic Cecropia and Musanga in the tree (Fig. 6). In the random sample of trees with C. sciadophylla and Musanga sister to Cecropia, the majority of ancestral reconstructions had probabilities of a myrmecophytic ancestor of Cecropia between 0.30 and 0.45. The distribution was approximately normal with the mean at a probability of 0.40 (Fig. 7). When C. sciadophylla and Musanga were arranged differently on the posterior trees, the majority of the probabilities of a myrmecophytic ancestor of Cecropia were between 0 and 0.25 with a mean probability of 0.35. The distribution was skewed to the left with a long tail (Fig. 7).
A comprehensive phylogenetic analysis of Cecropieae genera supports the emerging consensus based on molecular data that the tribe is monophyletic and that Poikilospermum belongs elsewhere in Urticaceae (Romaniuc-Neto 1999; Hadiah et al. 2008; Wu et al. 2013). Results of sampling three gene regions and all genera in the tribe agree with earlier studies in suggesting that the morphological similarities of Poikilospermum that led taxonomists to place it near Cecropieae are homoplasious. Morphology alone supported a sister relationship of Poikilospermum and Cecropieae (Fig. 2) whereas nuclear and chloroplast DNA sequences of Poikilospermum are more closely related to members of Urticeae (Figs. 3–5). We now turn our attention to understanding relationships among the remaining members of Cecropieae and their bearing on the evolution of myrmecophytism in this clade.
Molecular data supported the monophyly of three genera (Coussapoa, Myrianthus, and Pourouma), whereas Cecropia was rendered paraphyletic by the strongly supported position of Musanga (Fig. 5). This finding is not surprising given the morphological similarities of Cecropia and Musanga (Figs. 1, 2). Synonymizing Musanga, the smallest genus in the tribe, with the largest would restore the monophyly of Cecropia but Musanga leo-errerae Hauman & Léonard, has yet to be sequenced. Alternatively, C. sciadophylla could be transferred to Musanga but more complete sampling of Cecropia is needed to identify the most appropriate taxonomic change. DNA isolated from herbarium specimens has proven too degraded for analysis and it has not been possible to obtain new collections of the montane M. leo-errerae from East Zaire and Uganda. We predict that further study will support the synonmy of Musanga and the recognition of a more distributed Cecropia sensu lato encompassing both the Afrotropics and the Neotropics.
Musanga and Cecropia are ecologically similar lightdemanding pioneer trees in lowland forest succession with highly similar vegetative and reproductive characteristics (Table 1). Among the only distinguishing features are antassociated traits such as Müllerian bodies and trichillia that are present in most Cecropia species but are absent in Musanga. It is noteworthy that the only strongly supported intergeneric relationship in the three-gene phylogenetic analysis involved non-myrmecophytic C. sciadophylla and M. cecropioides. Most non-myrmecophytic Cecropia species are Andean high-altitude specialists occupying habitats where ants are either rare or absent (Latteman et al. 2014), whereas C. sciadophylla is a lowland species that often occurs in sympatry with other ant-associated Cecropia species. Another feature distinguishing Musanga from all but one Cecropia species is the absence of a spathe enclosing the infloresences. Cecropia hololeuca is a non-myrmecophyte that shares with Musanga a reduced and caducous bract in place of the spathe. These characters suggest that C. hololeuca could belong to the same non-myrmecophytic clade as C. sciadophylla and M. cecropioides. Janzen and McKey (1977) suggested that Musanga lost the ant association during migration from the Neotropics to Africa. However, considering that C. sciadophylla and M. cecropioides lack trichilia and Müllerian bodies, our results suggest that they shared a non-myrmecophytic common ancestor whose descendants dispersed across the Atlantic Ocean in one direction or the other. These alternative biogeographic scenarios for the evolution of Cecropia s. l. can be evaluated to some extent in the broader context of Cecropieae phylogeny and the fossil record.
Our study resolved the sister group to Cecropia s. l. with strong support from three gene regions for a clade including Pourouma, Coussapoa, and Myrianthus. The division of this group into a Neotropical clade consisting of Pourouma and Coussapoa and an Afrotropical clade (Myrianthus), together with the paleobotanical record, favors a Neotropical origin for the tribe followed by an ancient migration to Africa in the case of Myrianthus and a more recent migration in the case of Musanga. A Colombian fossil flora of the Maestrichtian Epoch with leaves that resemble Cecropieae suggests that the ancestors of the tribe were present in South America at least 65 million years before present (Burnham and Johnson 2004). Macrophyllous fossils from South American deposits dated 10–13 million years before present (Burnham and Graham 1999) have been unambiguously assigned to Coussapoa and provide further evidence that Cecropiaeae had already diversified in the Neotropics by at least the mid-Miocene. The alternative scenario, that Cecropiaeae were broadly distributed in Gondwana prior to the rifting of South America from Africa, is inconsistent with the low level of DNA sequence divergence observed in the group and relatively recent fossil-calibrated estimates of divergence time (Zerega et al. 2005).
The relatively large number of myrmecophytic Cecropia species (∼70) compared to a few species of non-myrmecophytic Musanga and Cecropia is consistent with the notion that mutualism could be associated with adaptive diversification (Weiblen and Treiber 2015). Mutualism may have enabled Cecropia populations to expand or occupy new niches if, for example, contributions of nitrogen from ants allowed Cecropia to thrive in nitrogen-limited environments (Sagers et al. 2000). Large population size would reduce extinction risk and niche expansion might have led to speciation. Defensive mutualisms involving ants and plants also appear to be accompanied by accelerated diversification rates (Weber and Agrawal 2014) and plant traits that reward ants have been regarded as key innovations (Lengyel et al. 2009). Understanding whether this is the case in Cecropia s. l. will require a more fully resolved phylogeny and more thorough sampling of myrmecophytes and non-myrmecophytes than in our study.
We encountered unexpectedly low levels of DNA sequence variation among Cecropia species (e.g. 0.8% and 0.13% phylogenetically informative characters for EPIC and ndhF, respectively) such that virtually no relationships within the genus were either resolved or supported in this study. Additional sequencing of Cecropia using more variable molecular markers such as the ribosomal internal transcribed spacer region (ITS) and glyceraldehyde-3-phosphate dehydrogenase (G3PDH) has yielded the same result (Treiber unpublished data). The difficulty of phylogenetic inference in Cecropia might be attributed to either a slow rate of molecular evolution or to a long history of hybridization and introgression (Xu 2000; Machado and Hey 2003). Next generation sequencing techniques can provide the quantity of data necessary to evaluate these alternatives and to resolve recalcitrant clades. Recent phylogenetic studies employing restriction-site associated DNA (RAD) sequence data (Baird et al. 2008; Emerson et al. 2010; Wagner et al. 2012; Eaton and Ree 2013) suggest that this technique holds promise for Cecropia. Rubin et al. (2012) found that RAD sequencing for phylogenetic analysis worked best for diploid species that diverged relatively recently (<60 Mya) which is the case for Cecropia.
Preliminary insights on relationships within Cecropia s. l. illustrate the challenges associated with ancestral state reconstruction (Fig. 6). The probability of myrmecophytism having originated in the common ancestor of the group varied depending on the position of non-myrmecophytic species relative to myrmecophytes. Most trees drawn from the Bayesian posterior distribution (97%) included the non-myrmecophytic clade (C. sciadophylla and Musanga) as sister to the rest of Cecropia where the probability of a myrmecophytic common ancestor ranged from 0.01–0.95 depending on the position of the other non-myrmecophytic Cecropia species (Figs. 6A, 6B, 7). The distribution of probabilities of a myrmecophytic ancestor for the sampled trees, when C. sciadophylla and Musanga were sister to Cecropia, was approximately normal with the mean at 0.40. When non-myrmecophytic C. sciadophylla and Musanga were embedded elsewhere, as was the case for 3% of the trees in the posterior distribution, the probability of a myrmecophytic common ancestor ranges from 0–0.99 (Figs. 6C, 7). The distribution of probabilities on the trees sampled for the different topologies was skewed towards zero but was fairly flat. Likely this is due to the number of different ways the five non-myrmecophytic samples may be embedded in the clade when C. sciadophylla and Musanga are not constrained as sister to the rest of the Cecropia clade. The ratio of myrmecophytes to non-myrmecophytes in our sample (8:5) compared to Cecropia as a whole (32:5) may also influence the ancestral reconstruction of ant association (Salisbury and Kim 2001). Our sample includes three of the eight non-myrmecophytic Cecropia species (Berg and Franco-Rosselli 2005) such that our estimates may be biased toward non-myrmecophytism (Fig. 7). Nonetheless, resolving relationships in Cecropieae and identifying the sister group to Cecropia are important first steps toward a robust classification of the clade and toward understanding the evolution of myrmecophytism in this group. These systematic findings provide a foundation for future investigation of the role that ant mutualism may have played in the radiation of Cecropieae and plant diversification in general. A phylogeny will also allow for reconstruction of ant-associated traits, such as domatia and Müllerian bodies, and tests for correlation with the origin of myrmecophytism, as well as other biologically interesting traits including the vertebrate-dispersed fruiting syndromes of Cecropieae.
We would like to thank C. Sagers for encouraging and supporting our work with Cecropia and K. Sousa for the illustration used in Fig. 1. We also thank J. Bevington, M. F. Torres, P. Barriga, and E. Kami for fresh plant material and the Paris Herbarium and the Missouri Botanical Garden Herbarium for access to herbarium specimens. We would like to acknowledge A. Guensberg and G. Kandlikar for laboratory assistance. P. C. Zalamea and M. F. Torres, who formed part of the Cecropia working group at Universidad de los Andes in Colombia, contributed in the field and in the herbarium to the great benefit of this study. This work was supported by a US National Science Foundation Grant to GW and SM (DEB-1132916). Support for ET was provided by the Bernard and Jean Phinney Graduate Fellowship in Plant Molecular Biology from the Department of Plant Biology, the Joyce Davenport Summer Fellowship the Bell Museum of Natural History, and the College of Biological Sciences Excellence Fellowship from the University of Minnesota. GW was supported by the Sanford Chair at Universidad de los Andes. AG was supported by CAPES, CNPq, FAPESP, and the Programa de Pos Graduação do Instituto de Botanica de Sao Paulo.