Reconstructing the origin and elaboration of insect-trapping inflorescences in the Araceae

Abstract

Premise of the study

Floral traps are among the most sophisticated devices that have evolved in angiosperms in the context of pollination, but the evolution of trap pollination has not yet been studied in a phylogenetic context. We aim to determine the evolutionary history of morphological traits that facilitate trap pollination and to elucidate the impact of pollinators on the evolution of inflorescence traps in the family Araceae.

Methods

Inflorescence morphology was investigated to determine the presence of trapping devices and to classify functional types of traps. We inferred phylogenetic relationships in the family using maximum likelihood and Bayesian methods. Character evolution of trapping devices, trap types, and pollinator types was then assessed with maximum parsimony and Bayesian methods. We also tested for an association of trap pollination with specific pollinator types.

Key results

Inflorescence traps have evolved independently at least 10 times within the Araceae. Trapping devices were found in 27 genera. On the basis of different combinations of trapping devices, six functional types of traps were identified. Trap pollination in Araceae is correlated with pollination by flies.

Conclusions

Trap pollination in the Araceae is more common than was previously thought. Preadaptations such as papillate cells or elongated sterile flowers facilitated the evolution of inflorescence traps. In some clades, imperfect traps served as a precursor for the evolution of more elaborate traps. Traps that evolved in association with fly pollination were most probably derived from mutualistic ancestors, offering a brood-site to their pollinators.

Changes in flower morphology have been of key importance for the diversification of angiosperms (Friis et al., 2006; Endress, 2011). The modification of floral organs, their increasing synorganization, and the evolution of new floral structures have enabled adaptation to a wide array of pollinators (Claßen-Bockhoff et al., 2004; Whittall and Hodges, 2007; Alcantara and Lohmann, 2010). In some groups, the interplay of these processes led to the evolution of very complex flowers and inflorescences (Harris, 1999; Rudall and Bateman, 2002). The reconstruction of the evolutionary history of such morphological changes in a phylogenetic context allows a better understanding of the general patterns of plant–pollinator interactions (Fenster et al., 2004; DeWitt Smith, 2010).

Floral traps (“Kesselfallenblumen” sensu Vogel, 1965, 1999) are among the most sophisticated devices that have evolved in angiosperms in the context of pollination. Their key innovation is the formation of a chamber, which encloses the sexual organs. The inner epidermis of the chamber entrance is slippery, causing insects—commonly attracted by means of deception—to slip and to fall into the chamber. The slippery surface consists of downward-pointing papillae and/or is covered with epicuticular wax crystalloids (Vogel and Martens, 2000; Poppinga et al., 2010). In some floral traps, the exit can be blocked either by hairs or elongated sterile flowers (Knoll, 1926; Sakai, 2002; Coombs et al., 2011) or by active closure of the floral chamber (Armstrong 1979; Dakwale and Bhatnagar, 1985). Floral traps are almost always protogynous (Vogel, 1961; Dafni, 1984; Thien et al., 2009). Pollinators are arrested for a defined period of time during the pistillate phase and can escape only during or after pollen release. In most cases, the escape is facilitated by the wilting of the trapping structures, while in few taxa a secondary opening of the floral trap creates a new exit (Vogel, 1965). In addition to perfect traps just described, imperfect traps (so-called semitraps after Faegri and Van der Pijl, 1971) also exist. In imperfect traps, insects are forced to exit the flower via a predetermined route so that pollen is deposited on their body. However, they are not arrested for a defined period of time. Perfect floral traps have evolved in at least eight unrelated families, predominantly in the basal angiosperms and the monocots (Dafni, 1984; Thien et al., 2009; Urru et al., 2011). Well-known examples are Aristolochia (e.g., Sakai, 2002) and Arum (e.g., Knoll, 1926; Gibernau et al., 2004). Although a number of studies have shed light on the interactions of floral traps and their pollinators (Vogel 1961; Diaz and Kite, 2002; Bolin et al., 2009), how these complex traps have evolved from nontraps is still unknown.

The present study is the first to analyze the evolution of floral traps in a phylogenetic context. Our overall aim is to determine the evolutionary history of morphological traits that facilitate trap pollination and to elucidate the impact of pollinators on the evolution of traps. Our study system is the Araceae, a diverse family comprising over 3300 species in 126 genera (Boyce and Croat, 2012). The most characteristic feature in this family is the inflorescence, which consists of a thickened flower-bearing spike, called the spadix, and a single, usually conspicuous bract, called the spathe (Fig. 1A). The spathe, the spadix, and the flowers are subject to various modifications and increasing synorganization. While the spathe is often inconspicuous or simply expanded in basal clades such as Gymnostachyoideae and Orontioideae, it frequently surrounds the spadix and forms a chamber around the flowers in higher clades such as Lasioideae and Aroideae (note that terms such as basal or higher used in the text for taxa refer to the topology of the phylogenetic tree and do not indicate primitiveness/advancement of any given character; Crisp and Cook, 2005). Moreover, in the majority of clades the spadix is completely covered by bisexual flowers, while in the Aroideae the flowers are unisexual and arranged in distinct zones: the pistillate flowers are situated on the lower portion of the spadix and the staminate flowers on the upper portion, often separated by a zone of sterile flowers. In higher taxa of Aroideae such as Arum and allies the upper part of the spadix becomes sterile and serves as an osmophore (Fig. 1B). The whole family is protogynous. In clades with bisexual flowers, anthesis lasts for several days, while in monoecious taxa of Aroideae anthesis usually ceases after 1–2 d. Pollinators of Araceae are Diptera, Coleoptera, and Hymenoptera (Gibernau, 2011). Interactions include food reward, mating mutualism, nursery mutualism, and deception (Gibernau et al., 2010; Chartier, 2011). Trap pollination is known from several clades of the subfamily Aroideae and includes both perfect as well as imperfect traps (Ørgaard and Jacobsen, 1998; Vogel and Martens, 2000; Gibernau et al., 2004), but most of the genera of Araceae apparently have no traps. The diversity of inflorescence forms and the occurrence of different pollination syndromes make the Araceae an ideal object for studying the evolution of trap pollination.

Fig. 1.

Inflorescence morphology of selected Araceae. (A) Anthurium digitatum (Jacq.) Schott, spadix with bisexual flowers and an expanded spathe. (B) Arum nigrum Schott, spadix with unisexual and sterile flowers and a spathe separated into a tube and a blade. The front part of the spathe tube is removed for better visibility of the spadix; a = pistillate flowers; b = elongated sterile flowers; c = staminate flowers; d = sterile appendix. Scale bars = 1 cm.

In this paper, we specifically address the following questions: (1) In which taxa of Araceae do trapping devices occur and how did they evolve? (2) Did the different types of inflorescence traps evolve from a common trap-ancestor and did perfect traps evolve from imperfect traps? (3) Are traps associated with specific insect groups? To assess the frequency of trap pollination in the family, we examined the presence and structure of trapping devices in taxa from all major clades. In addition, we reconstructed the phylogeny of Araceae based on the molecular data of Cusimano et al. (2011), complemented by sequences of one additional taxon. Trapping structures, trap types, and pollinator types were then mapped onto the phylogeny and the ancestral states were reconstructed with maximum parsimony and Bayesian inference.

Materials and methods

Plant material

We studied taxa from all genera available covering all tribes of the family (sensu Mayo et al., 1997) except for the monotypic subfamily Gymnostachydoideae, for which no samples were available (and which with certainty does not form traps). Inflorescences were collected during anthesis in the field and in botanical gardens. The samples were stored in 70% alcohol as well as dried at room temperature. We also used material already stored in ethanol from various wet collections. Voucher specimen information is listed in Appendix 1.

Occurrence, structure, and function of traps

Trapping devices

Based on the trap characters defined by Vogel (1965, 1999), we examined the four morphological characters that—alone or in combination—are essential to trap and detain pollinators and thus allow us to infer the presence of a trap from the inflorescence morphology: (1) Spathe shape: We used the typification of spathe type by Grayum (1990). Additional information on missing or misclassified genera in Grayum’s study was taken from the more recent descriptions of spathe shape in Mayo et al. (1997). Grayum’s spathe types are: type 1 = unmodified, bractlike; type 2 = expanded and/or colored (including boat-shaped Monsteroideae), type 3 = enclosing spadix, i.e., spathe margins convolute at least in the lower part of spadix; type 4 = constricted, i.e., the spathe more or less completely surrounds the spadix, forming a basal chamber and an apical blade. Only taxa with spathe type 3 or 4 can form traps as the presence of an at least rudimentary chamber is essential for retaining insects. (2) Slippery surface: The presence of epicuticular wax crystalloids and/or downward pointing papillate cells on the epidermis of the spathe was studied with scanning electron microscopy (SEM) (JEOL JSM6390, Akishima, Japan). We investigated samples of 142 species in 76 genera. Samples were taken from all different regions of the organ after preliminary investigation under light microscope (Olympus BX50, Tokyo, Japan). Samples used for the assessment of cell shape were dehydrated in an increasing series of ethanol from 70% to 85% to 96% for 20 min in each solution and then transferred to acetone. Consecutively, samples were critical-point-dried and sputter-coated with gold and investigated with SEM. Samples used for the examination of epicuticular wax crystalloids only were air-dried before sputter-coating, because ethanol and heat would alter the crystal structure of the wax (Barthlott and Wollenweber, 1981). (3) Elongated sterile flowers: Information on the presence of elongated sterile flowers was taken from Mayo et al. (1997). (4) Temporary closure of the spathe during anthesis: Spathe movements in specimens cultivated in the Botanical Garden of Vienna were recorded during daily observations. We also used a Nikon Coolpix P 5000 camera (Tokyo, Japan) to take images automatically every 10 min. In addition, information on spathe movements was also taken from Mayo et al. (1997).

Functional types of traps

In our study, we define a genus as having a trap when the spathe shape is “enclosing the spadix” (type 3) or “constricted” (type 4) and when one or more of the aforementioned trapping devices were also present. For our results, we identify and classify the range of functional types of traps resulting from the different combinations of these trap characters. The functional types also relate to the mode of operation of traps based on the following stages (sensu Vogel, 1965): (1) mode of capture, (2) mode of retention, and (3) mode of release of pollinators.

Reconstruction of the evolutionary history

Molecular phylogeny

We reconstructed the phylogeny of the Araceae using the molecular matrix of Cusimano et al. (2011), which includes 113 genera of Araceae and three out-group taxa (Acorus, Hedyosmum, Tofieldia). The alignments of multiple chloroplast markers (rbcL, matK, partial trnK intron, partial tRNA-Leu gene, trnL-trnF spacer, and partial tRNA-Phe gene) were obtained from TreeBase (study 11083, tree Tr26254). We added sequences for one taxon, namely Colocasia gigantea (Blume) Hook.f., which had been shown to be more closely related to Alocasia than to the other taxa of Colocasia (Renner and Zhang, 2004; Nauheimer et al., 2012). Sequences of Colocasia gigantea were downloaded from GenBank (EU193194.1, EU886581.1, EU193409.1, EU193321.1) (Cusimano et al., 2008, 2010; Mansion et al., 2008) and aligned manually. The new matrix of the combined regions consisted of 117 taxa and 4498 aligned characters.

Sequence data were analyzed with maximum likelihood and Bayesian methods following Cusimano et al. (2011). The best fitting model of evolution was determined as GTR + Γ by the Akaike information criterion (Akaike, 1974) as implemented in the program jModelTest v0.1.1 (Posada, 2008). For maximum likelihood analysis, we used the software RAxML 7.2.8 (Stamatakis et al., 2008) available through CIPRES Science Gateway (Miller et al., 2010). RAxML uses the GTRCAT approximation of the GTR + Γ model, with the gamma shape parameter having 25 rate categories. Bootstrap values were obtained by running 1000 replicates. The Bayesian analysis was run with the program MrBayes (Huelsenbeck and Ronquist, 2001; Ronquist and Huelsenbeck, 2003), available at the Bioportal cluster (http://www.bioportal.uio.no). We performed four runs of eight million generations, with trees sampled every 500th generation. The convergence diagnostic in MrBayes was used to assess the convergence of all runs. For each run, the first 25% of the resulting 40 000 trees were discarded as burnin. For consecutive analyses, we sampled 10 000 Bayesian trees from two different runs. A 50% majority-rule consensus tree was reconstructed for which polytomies were randomly resolved, and a length of 10⁻⁷ was assigned to branches with zero or negative length using the software Mesquite 2.0 (Maddison and Maddison, 2007).

Character evolution

Character states of all trapping devices and the functional types of traps were mapped onto the Bayesian 50% majority-rule consensus tree, and ancestral states were reconstructed applying maximum parsimony (MP) in Mesquite and Bayesian analysis in the software SIMMAP 1.5 (Bollback, 2006). The outgroup taxa (i.e., Acorus, Hedyosmum, Tofieldia) were excluded from the analyses. Molecular rates of evolution are correlated with generation time (Smith and Donoghue, 2008). Since we have a wide range of generation times in Araceae, we chose to use the phylogram instead of an ultrametric tree for both analyses. In SIMMAP, we used all 10 000 Bayesian trees to calculate posterior probabilities (PP) for ancestral states, thus taking into account phylogenetic uncertainties. To assess how prior choice may influence the posterior results (Schultz and Churchill, 1999), we used two different sets of priors for all characters studied. One set of priors was calculated with the MCMC approach offered in the software, the second set consisted of the program’s default priors. The advantage of the MCMC approach is that overall rate values are sampled and the best fitting gamma distribution can be found, instead of guessing and trying a large number of different priors. However, one has to keep in mind that the results of ancestral state reconstructions always depend on the underlying assumptions and that inferences may fail if the model applied is unrealistic (see Crisp and Cook, 2005). Results are shown for calculations with the MCMC prior if not stated otherwise. As SIMMAP does not allow polymorphic character states, we coded them as ambiguous. For the mapping of the character trap type, we used an additional approach in Mesquite: it can be argued that transitions between character states are more likely to be imbalanced in complex characters due to asymmetric gain-loss probabilities (for a detailed discussion, see Kohn et al., 1997; Omland, 1999). Thus, transitions between different types of traps may occur with higher probability than transitions between nontraps and traps. Therefore, we made a step matrix for the MP reconstruction in Mesquite, where each transition between two trap types costs one step, while a transition between a trap and a nontrap costs two steps. The average number of changes between the different types of traps (including nontraps) was estimated with the “summarize changes” option in Mesquite for 10 000 Bayesian trees reconstructed with the MP method.

Association between inflorescence traps and pollinators

Pollinators

Information on pollinator types in Araceae was taken from reviews by Gibernau and his coworkers (Gibernau, 2003, 2011; Gibernau et al., 2010). In addition, information on pollination of Dracunculus vulgaris Schott in H. W. Schott & S. L. Endlicher was taken from Schmucker (1930) and Meeuse and Hatch (1960). Lasia spinosa (L.) Thwaites has been observed to be visited by flies in Xishuang-banna Tropical Botanical Garden, Yunnan, China (Yin J. T., personal communication). Pollinators were classified into four categories: Diptera, Coleoptera, Hymenoptera, and generalist (i.e., more than one type of pollinators). Ancestral states of pollinator type were reconstructed using the methods described above.

Character association

We tested for correlated evolution of trap pollination with pollinator type using the programs SIMMAP 1.5 and MacClade 4 (Maddison and Maddison, 2000). The different trap types were summed up under the character trap presence with the states trap and nontrap. For all correlations, we used the Bayesian 50% majority-rule consensus tree. All taxa with unknown character states were removed, resulting in a matrix with 54 taxa. SIMMAP allows multiple comparisons of characters with an unlimited number of character states. The software calculates the time that characters spend in particular states along the tree as a measure for association. The observed distribution of the character states is then compared to a predicted distribution to assign a P value. We ran 1000 simulations and drew 500 predictive samples to calculate the P value. To check whether prior choice influences the results, we made all calculations using the priors calculated with the MCMC approach offered by the software, as well as with the default priors. The Character Correlation Test (CCT) in MacClade 4 only allows two binary characters to be correlated. Therefore, we coded pollinators as Diptera/other or Coleoptera/other, and correlated each of these two characters schemes with trap presence. The CCT tests whether changes in the dependent character (i.e., trap presence) are more concentrated than expected on those branches that have a particular state in the independent character (i.e., fly/ beetle pollination). We used ACCTRAN and DELTRAN options to resolve equivocal reconstructions of ancestral character states and applied MINSTATE and MAXSTATE reconstructions for the calculation of correlated evolution. For each run, 100 000 simulations were performed.

Results

Occurrence of traps

Trapping devices

The coding of all characters studied is shown in Fig. 2. Character sampling covers the majority of clades but is not complete. The reader therefore must be aware that inferences may vary if missing taxa are added or scored. The most common spathe shapes found across the 114 genera were type 3 (spathe enclosing spadix) (found in 37genera) and type 4 (spathe constricted) (36 genera). Both these types were especially abundant in the Aroideae subfamily. Unmodified spathes (type 1) were most common in Gymnostachydoideae, Orontioideae, and Lemnoideae (14 genera). Expanded/colored spathes (type 2) were found in 20 genera, 11 of which belong to the subfamilies Monsteroideae and Pothoideae. In addition, seven genera contain species with different spathe shapes.

Fig. 2.

Ancestral state reconstructions of trap pollination in Araceae. Colors on lines indicate reconstruction of trap types (A) with maximum parsimony in Mesquite 2.0. In cases where lines have more than one color, ancestral states could not be resolved unambiguously. Pie charts on the nodes display the posterior probabilites of trap types calculated with Bayesian inference in SIMMAP 1.5. Pie charts below the nodes display the posterior probabilites of pollinator types (B). Arrows point at the earliest appearance of the four types of spathe shape (C). Bars indicate the appearance of characters D-G along the phylogeny. Coding for all characters is shown at the right. Node numbers are referred to in the text.

Of 76 genera studied under SEM, 31 possessed either epicuticular wax crystalloids, papillate cells, or both on their adaxial spathe surface (Fig. 3). Papillate cells in Amorphophallus, Colocasia, Pseudodracontium, Stylochaeton, Ariopsis, Colocasia gigantea, Remusatia, and Zantedeschia were not downward pointing but orientated horizontally. Because the latter four taxa did not have any additional trap characters, they were coded as equivocal for trap type. Epicuticular wax crystalloids formed platelets (Fig. 3A, F), tubules (Fig. 3B), threads (Fig. 3B), or branched rodlets (Fig. 3.C). Downward pointing papillae were found in 16 genera (Fig. 3C–E). In Lasioideae, downward pointing papillae usually formed imbricate rows like roof tiles (Fig. 3E). Moreover, several lasioids had an additional epicuticular wax layer (Fig. 3C). In Pycnospatha, the cells were flattened, and papillae were no longer recognizable. In this case, the function of a slippery surface was completely transferred to the epicuticular wax crystalloids. In the subfamily Aroideae, downward pointing cells were not fused (Fig. 3D). In contrast with slippery surfaces in the Lasioideae, the co-occurrence of an epicuticular wax layer was found to be rare in the Aroideae, except for a few taxa such as Amorphophallus (Fig. 3A), where straight papillate cells and epicuticular wax crystalloids are present. Moreover, wax crystalloids were also detected on the sterile appendix of the spadix in several species of Amor-phophallus. In several taxa, cuticular folds were present on papillate cells (Fig. 3E). These folds also occurred in nontraps with tabular or convex epidermal cells, for example, in several genera of tribe Spathicarpeae (Fig. 2, node 107). A detailed description of cell shape, structure of wax crystalloids, and presence of cuticular folds in all taxa studied is presented in Appendix S1 (see Supplemental Data with the online version of this article).

Fig. 3.

Slippery surfaces on the adaxial spathe epidermis in Araceae. (A) Wax platelets (Steudnera kerrii Gagnep.). (B) Wax tubules and threads (Amorphophallus taurostigma Ittenbach, Hett. & Bogner). (C) Imbricate downward pointing papillate cells and branched wax rodlets [Urospatha sagittifolia (Rudge) Schott]. (D) Downward pointing papillate cells (Helicodiceros muscivorus L.f.). (E) Imbricate downward-pointing papillate cells with cuticular folds (Dracontium asperum K. Koch). (F) Wax platelets on perpendicular papillate cells (Stylochaeton cf. hypogaeus Lepr.). Note: Cells in samples A–C and F have shrunk due to drying at room temperature. Scale bars = 10 µm.

Elongated sterile flowers were found in 12 genera, eight of which belong to the tribe Areae. Temporary closure of the spathe during anthesis occurred in seven genera. In Colocasia (Fig. 4A–C) and Schismatoglottis, the entire spathe blade closes. In Sauromatum, Theriophonum, and Typhonium (Fig. 4D–F), only the constriction closes, thereby secluding the basal chamber.

Fig. 4.

Spathe closure during anthesis. (A–C) Colocasia fontanesii Schott. (A) The spathe blade opens a narrow slit (arrowhead) during the pistillate phase. (B) The spathe blade closes at the end of the pistillate phase. (C) The spathe blade reopens and reflexes during the staminate phase. The constriction above the spathe chamber is now closed. (D–F) Typhonium trilobatum (L.) Schott. (D) The spathe constriction above the floral chamber opens at the beginning of the pistillate phase and spadix tilts forward. (E) The constriction closes at the end of the pistillate phase; the spadix is erect. The color of the spathe blade gradually turns from red to brown. (F) The constriction reopens during the staminate phase, and the spadix tilts forward again. Scale bars = 1 cm.

Functional types of traps

On the basis of the presence and combination of trapping devices, we could identify six types of traps (Fig. 5). (1) In the Schismatoglottisype (Fig. 5A), insects are trapped by temporary closure of the spathe blade, thereby enclosing the whole spadix. A slippery surface is not present. The spathe is constricted, forming a chamber around the pistillate flowers at the lower part of the spadix. The spathe blade usually opens a small slit only during pistillate phase. The constriction closes after insects have left the chamber and moved to the upper part of the inflorescence. In the subsequent staminate phase, the spathe blade expands and often bends back abruptly, thus exposing the upper part of the spadix. The spadix itself is composed of a sterile zone located between pistillate and staminate flowers. In several taxa, a sterile appendix above the staminate flowers is also present. Anthesis lasts for about 24 h. The Schismatoglottis type was found in only two genera of Aroideae, Schismatoglottis and Colocasia. (2) In the Arisarum type (Fig. 5B), no spathe closure occurs, but slippery surfaces (wax, downward pointing papillae) are present on the spathe. Spathe shape usually is an “enclosing-spadix” type. This trap type occurs in the subfamilies Lasioideae and Aroideae. In the Arisarum type traps of the Lasioideae, spathe margins are convolute only in the lower part of the spathe, while in the Aroideae the floral chamber makes up more than half of the spathe and encloses at least a part of the staminate section of the spadix. Anthesis lasts for several days to weeks. The spadix does not contain any elongated sterile flowers, which might serve as a barrier and the spathe is usually not constricted. Therefore, insects will glide down into the lower part of the spathe but are not arrested and can escape by climbing the spadix and/or flying out of the chamber. Thus, traps of the Arisarum type represent imperfect traps. (3) Traps of the Zomicarpa-type (Fig. 5C) are similar in shape to the Arisarum type, but the spathe margins are always convolute to the upper third of the spathe, and the fertile part of the spadix is completely hidden inside. The entry to the floral chamber often is masked by a hooded spathe blade. This type of trap occurs in subfamily Lasioideae as well as in Aroideae tribe Arisaemateae (Fig. 2, node 209) and the genus Zomicarpa. Slippery surfaces often consist of wax crystalloids, which can also be present on the sterile appendix in Arisaema. In several taxa, the epidermis of the spathe can also consist of downward pointing papillate cells. Insects cannot escape until the spathe margin bulges out at the lower spathe or opens completely and builds a secondary exit. Arisaema is unique in being dioecious (Vogel and Martens, 2000). Only the male inflorescences provide an exit for insects. The female inflorescences remain closed, and the captured insects cannot escape. Anthesis usually lasts for several days to weeks. (4) Traps of the Typhonium type (Fig. 5D) are found in the tribes Cryptocoryneae (Fig. 2, node 134) and Areae (Fig. 2, node 212). Here, slippery surfaces are made up by papillate cells. The spathe closes temporarily. There are two means by which the floral chamber is secluded. In Cryptocoryneae an extension of the spathe margin occludes the floral chamber, while in Areae a twist of the spathe causes the closure of the constriction between the floral chamber and the blade. All taxa of the Typhonium type have monoecious inflorescences. In most taxa, the floral chamber also encloses the staminate flowers, while in Typhonium they are situated above the constriction. Therefore, pollen does not fall into the floral chamber but is deposited on the constriction. After the spathe has opened again and the slippery surface ceased to be slippery, insects can escape. Anthesis usually lasts for 24 h. (5) In the Arum type (Fig. 5E), which is restricted to four taxa of the tribe Areae, traps do not close their constriction. During anthesis, escape is prevented by the presence of downward pointing papillae on the spathe and slippery elongated sterile flowers on the spadix. After anthesis, these parts wither, and the insects can leave the trap through climbing. Elongated sterile flowers can occur in one or two whorls, and their shape is subulate to filiform. Pistillate and staminate flowers are enclosed by the floral chamber. Anthesis usually lasts for 24 h. (6) Stylochaeton appears to have a unique trapping mode (Fig. 5F). The gliding surface consists of straight papillate cells and an epicuticular wax layer. At the beginning of anthesis, the spadix is hidden inside the spathe. When pollen is released, the spadix starts to grow above the spathe chamber, forming a ladder that presumably facilitates the escape of insects. Anthesis lasts for one to a few days.

Fig. 5.

Functional types of traps in Araceae. (A) Schismatoglottis type. The insects are retained by spathe movements; slippery surfaces are absent. (B) Arisarum type. An imperfect trap with a slippery spathe surface. Insects slip and fall into the spathe chamber but can escape unhampered by climbing the spadix or flying off. (C) Zomicarpa type. Insects are trapped inside the inflorescence by slippery surfaces and are released through a secondary exit formed by a movement of the spathe. (D) Typhonium type. Insects glide down slippery surfaces and are retained in the floral chamber by closure of the spathe constriction. During the staminate phase, the constriction reopens, and slippery surfaces cease to be slippery. (E) Arum type. The insects are trapped by slippery spathe surfaces and sterile flowers on the spadix that partially occlude the spathe chamber. Insect release is facilitated by withering of the slippery organs. (F) Stylochaeton type. Insects are trapped by slippery spathe surfaces. In the pistillate phase, the spadix is enclosed in the spathe chamber. During the staminate phase, the spadix grows out of the chamber, and insects can escape via climbing. Gender symbols indicate pistillate and staminate phase of anthesis. Black arrows indicate arrival and departure of pollinators. Arrowheads indicate closure of the spathe constriction. The insect symbol indicates the pollinator’s residence during arrest. Asterisks indicate the presence of an intact slippery surface; crosses indicate that the slippery surface has withered and ceased to be functional.

Reconstruction of the evolutionary history of traps

Phylogeny

The topologies of the ML and Bayesian analyses proved consistent with those of Cusimano et al. (2011). The additional taxon Colocasia gigantea is grouped with Alocasia. The branch is strongly supported with a bootstrap support of 100 and a Bayesian posterior probability of 1 (Fig. 2, node 205).

Character evolution

Ancestral state reconstructions of trap type and selected transitions in other trap characters are shown in Fig. 2, the 50% majority-rule consensus tree of 10000 Bayesian trees sampled from two runs. Detailed ancestral state reconstructions of all trap characters not displayed in the main figures are shown in Appendix S2–S6 (see online Supplemental Data). Results of MP and Bayesian approach were consistent overall. Different reconstructions were found in the character trap type in the common ancestors of four clades (Fig. 2, nodes 53, 134, 154, 199). In the MP analysis, these nodes were reconstructed as having traps because the step matrix favored transitions between different trap types. In contrast, in the Bayesian analysis, these nodes were reconstructed as nontraps from which different types of traps were derived. Thus, the Bayesian approach shows a higher number of changes between traps and nontraps.

Two ways of handling the choice of prior in Bayesian analysis did not affect the reconstruction of character history except for the character spathe shape. The parameters for the default Γ-prior for multistate characters were α = 1.25, β = 0.25, and α = 1.00 for the default B-prior for two-state characters. The priors calculated with the MCMC-approach for the various characters are: trap type (Γ: α = 8.58, β = 0.24), spathe shape (Γ: α = 16.62, β = 0.37), epicuticular wax (B: α = 11.55), papillate cells (B: α = 11.28), elongated sterile flowers (B: α = 2.65), and spathe closure (B: α = 5.05). For spathe shape, the calculations with the default prior yielded results more similar to the MP reconstruction than the calculations with the MCMC prior. The common ancestor of Araceae (Fig. 2, node 2) was most likely type 1 (unmodified) (PP = 48%) with the default prior (followed by enclosing spadix, PP = 39%), while it was most likely type 3 (enclosing spadix) with the MCMC prior (PP = 46%) (followed by constricted, PP = 39%). The majority of nodes did not change in their reconstructed ancestral state with a different choice of prior. In the MP approach, node 2 is reconstructed as spathe type 1 in accordance with the Bayesian reconstruction using the default prior (Appendix S2, see online Supplemental Data). Apart from Lasioideae and Aroideae, where spathe shape 3 is more common, it is only present in one extant taxon of subfamily Orontioideae (Fig. 2, node 5) and two taxa of subfamily Monsteroideae, while spathe shape 1 is more common in the latter clades. We present the results for spathe shape calculated with the default prior in the Bayesian analysis in Fig. 2. In all reconstructions, inflorescences that have at least a rudimentary floral chamber had already evolved very early in the history of Araceae, possibly in the branch preceding node 10 (PP = 43%). Spathe shape 4 (constricted) only occurs in subfamily Aroideae. We found that it evolved several times within basal clades of the subfamily and formed the ancestral state of a large clade including all remaining taxa of Aroideae (Fig. 2, node 130). Within this clade, spathe shape 4 was lost several times. Epicuticular wax crystalloids, papillate cells, elongated sterile flowers, and spathe closure during anthesis have evolved repeatedly in various clades (Appendix S3–S6). In Lasioideae, Stylochaeton and Arisarum the former two traits occur simultaneously. In most cases, they are associated with trap pollination. Rhaphido-phora and Scindapsus had the only two nontraps with a pronounced epicuticular wax layer in our study. However, the wax layer does not contain three-dimensional crystalloids, but forms an irregular crust. Furthermore, their spathe shape is rather boat-shaped without convolute spathe margins. Elongated sterile flowers occur in Amorphophallus, Arisaema, Dracontium, Bucephalandra, and in the tribe Areae.

Inflorescence traps have evolved at least 10 times within the family (Fig. 2). The common ancestor of Araceae does not have a trap (PP = 100%). Moreover, traps do not occur in any genus of the basal clades. Among the basal subfamilies with inflorescence having a lower degree of synorganization, Lasioideae is the only clade where traps have been found. Most traps occur in the Aroideae where at least seven independent events have led to the formation of traps in at least 19 genera. With the exception of the Arum type and the Stylochaeton type, all functional types of traps have evolved in more than one clade. The Stylochaeton type is restricted to a single genus and has evolved from a nontrap. The Arisarum and the Zomicarpa type are the most widespread types, being also present in Lasioideae. In this subfamily, the latter type is derived from the former in Dracontioides. The trap type of the common ancestor of all Lasioideae (Fig. 2, node 53) could not be resolved due to the different states in Lasia and the core lasioids. In the Bayesian analysis, it is inferred to be a nontrap (PP = 88%), implying that traps have evolved twice in Lasioideae, i.e., once in Lasia and once separately in the remaining clade. For MP, the node is reconstructed as a trap (equivocal for the Arisarum type and the Zomicarpa type). In the Areae clade, the Arum type is derived from the Typhonium type. The common ancestor of the clades Areae and Arisaemateae cannot be resolved unambiguously. In the Bayesian analysis, the Zomicarpa type has the highest probability (PP = 39%) followed by the Typhonium type (PP = 23%), while in MP it is equivocal (Zomicarpa or Typhonium type).

The average number of changes between traps and nontraps (mean ± SD) over 10 000 trees was 9.7 ± 0.73 changes from nontraps to traps and 2.8 (±0.99) reversals. Changes occurred most often from nontraps to imperfect traps of the Arisarum type (4.20 ± 0.41) and from the latter to the Zomicarpa type (1.34 ± 0.35). In all other transitions, ≤1.0 changes occurred.

Association between trap pollination and pollinators

Pollinators

Prior choice in the Bayesian approach had an impact on the ancestral state reconstruction of pollinator types. In the reconstruction using the MCMC prior (Γ: α = 2.86, β = 0.01), pollination by Diptera prevailed in the majority of clades, even if the extant taxa were not pollinated by flies (Appendix S7). The reconstruction using the default prior was considerably different, as fly pollination was less common except for subfamily Aroideae. Here we focus on discussing the ancestral state reconstruction of the Bayesian analysis calculated with the default prior in Fig. 2 since it is closer to the results of MP analysis in Mesquite. It was not possible to reconstruct the pollinator type of the common ancestor of Araceae unambiguously. In the Bayesian analysis, the most probable common ancestor was Diptera with a posterior probability of 46%, followed by generalist pollinators (30%) and Coleoptera (24%). The reconstruction with MP was equivocal, too (Appendix S7). Bee pollination is restricted to the subfamilies Pothoideae and Monsteroideae (Fig. 2, node 21). The common ancestor of subfamily Aroideae was most probably pollinated by beetles (PP = 60%). Subsequently, a change from beetle to fly pollination occurred in the branch leading to node 130 (Fig. 2). With the exception of tribes Caladieae (Coleoptera) and Areae (generalist pollination by flies and beetles), all higher clades of Aroideae have Diptera as the ancestral state. Of 10 clades containing traps, six clades most probably had a fly-pollinated ancestor, whereas beetle pollination was ancestral only once (i.e., Zomicarpa). In two clades, the ancestral state was ambiguous, and in one clade, pollination was probably achieved by more than one type of pollinator.

Character association

Results of the correlation analyses are shown in Table 1. Regardless of the method of reconstruction applied, pollination by Diptera was significantly correlated with trap pollination in CCT. In contrast, there was no significant correlation between Coleoptera and trap pollination. Likewise, regardless of prior choice considered here, SIMMAP found a correlation between trap pollination and Diptera, although the correlation became nonsignificant after application of Bonferroni-correction for multiple comparisons. Pollination by Coleoptera, Hymenoptera, or generalists was never correlated with trap pollination.

Table 1.

P values for correlation of trap pollination and pollinator type in Araceae calculated with CCT in MacClade 4 and the character association test in SIMMAP 1.5. P values for results in SIMMAP before Bonferroni correction are in parentheses.

Test	Diptera	Coleoptera	Hymenoptera	Generalist
CCT ACCTRAN	<0.01	0.22	—	—
CCT DELTRAN	<0.01	0.17	—	—
SIMMAP	ns (0.04)	ns (−0.09)	ns (−0.16)	ns (0.46)

Discussion

The evolution of floral traps depends on the presence of several morphological traits that facilitate the capture and retention of pollinators (Vogel, 1965). To understand the drivers for the evolution of trap pollination in the Araceae, we studied (1) the occurrence and function of trapping devices, (2) the emergence of different types of traps, and (3) the association between traps and the pollinating fauna.

Occurrence of traps and the origin of trapping devices

This study demonstrates that trap pollination is more widespread in Araceae than was previously thought. Inflorescence traps are present in at least 27 genera. We found that they are not restricted to several clades of the subfamily Aroideae, but also occur in Stylochaeton (Stylochaeton clade sensu Cusimano et al., 2011) as well as in several genera of the subfamily Lasioideae.

The precondition for the evolution of traps was the presence of a floral chamber formed by the spathe. Although we cannot completely exclude other possibilities, our ancestral state reconstructions indicate that the common ancestor of Araceae most likely had a bract-like spathe. Subsequently, a floral chamber had already evolved in the early history of the family. Nevertheless, this key innovation was not immediately followed by the evolution of traps. Therefore, it is probable that the spathe chamber is a preadaptation that originally served another function. As a bract, the spathe surrounds and thus protects the developing inflorescence. This was most likely its ancestral—and only—function (Grayum, 1990). In extant aroids, there are many further functions. In several taxa, the spathe is expanded and colored, an aid in attracting pollinators (Grayum, 1990; Kraemer and Schmitt, 1999). In other taxa, the spathe base remains furled round the flowers to form a floral chamber throughout flowering and seed set (e.g., Alocasia, Caladium, Dieffenbachia, Philodendron). Here, it often serves as a mating chamber or brood site (Gibernau et al., 2000; Miyake and Yafuso, 2005; Maia and Schlindwein, 2006). Provision of such rewards is essential for pollination success, as these guarantee that the insects will stay in the inflorescence until the staminate phase. Young (1986) showed that beetles that fed on sterile flowers in the spathe chamber of Dieffenbachia longispatha left the inflorescence before pollen-shedding when these food bodies were removed. The evolution of the ancestral spathe chamber thus probably included functions found in extant species such as shelter, food rewards, and/or a mating site for its pollinators (Chartier, 2011). Plant–pollinator interactions in such rewarding taxa differ fundamentally from true traps, and one should be cautious to deduce the presence of trap pollination from the shape of the inflorescence alone, because trapping depends on further trapping devices.

The most common device for trapping insects is a slippery plant surface. Such surfaces are composed of an epicuticular layer of wax crystalloids and downward pointing papillate cells (Poppinga et al., 2010). In Araceae, both traits have evolved multiple times, in some cases concurrently (Fig. 2). We found slippery surfaces with wax crystalloids of various shapes ranging from scale-like platelets to long threads. Through their three-dimensional structure, the crystalloids reduce the surface to which insect’s legs can attach and thus impede adhesion. Moreover, the crystalloids also can break off and stick to the insect’s adhesive pads (Gaume et al., 2004). Such wax crystalloids can be found throughout the angiosperms (Barthlott et al., 1998). They have evolved repeatedly in various contexts of plant–pollinator interactions (Eigenbrode, 2004; Gaume et al., 2004; Quek et al., 2004) and are also found on the foliage leaves of some Araceae (Koch et al., 2008). Because wax crystalloids are easily formed and are absent in many taxa it is most likely that they evolved de novo in the context of trap pollination.

Downward pointing papillae not only function as slippery surfaces because of their shape, but also secrete oil, which increases slipperiness (Knoll, 1926; Yadav, 1998). After pollen release, they often collapse, thus facilitating the escape of pollinators (Dakwale and Bhatnagar, 1982; Lack and Diaz, 1991; Bermadinger-Stabentheiner and Stabentheiner, 1995). We found several aroid taxa with papillate cells on the adaxial surface of the spathe, which did not point downward but projected perpendicularly to the spathe surface. Whether this kind of “straight” papillae can also form a slippery surface is not clear. Ivancic et al. (2004) mention that the (papillate) spathe surface of Colocasia esculenta was slippery for flies. However, according to our own field observations in the same species as well as in Colocasia fontanesii, drosophilid pollinators can walk along the spathe (Bröderbauer et al., unpublished manuscript). “Straight” papillate cells also occur in Zantedeschia. In Zantedeschia var. elliotiana, we observed (in the Botanical Garden of Vienna) trapped wild bees that were unable to climb the lower papillate portion of the inner spathe (Bröderbauer, unpublished manuscript). However, experimental proof that such cells can form a slippery surface is still missing. If they produce oil they might easily be slippery without pointing downward.

Papillate cells also might serve another function. While the spadix is the most common organ for scent-production, the spathe has also been shown to be an osmophore in some aroid taxa such as Arisaema, Cryptocoryne, and Dracontium (Vogel, 1963; Mayo et al., 1997; Zhu and Croat, 2004). During our study, we found that most of the papillate slippery surfaces also emitted faecal odours, often similar to those of the spadix and changing during the course of anthesis. In fact, osmophoric plant surfaces reported by other researchers (Vogel 1963; Stpiczyńska, 2001; García et al., 2007; Płachno et al., 2010) often resemble the “straight” papillate cells in Araceae. Whether (“straight”) osmophoric papillae are ancestral and subsequently changed their function toward slippery surfaces has yet to be demonstrated. However, in the ‘Pistia-clade’ (sensu Renner and Zhang, 2004; Fig. 2, node 194), which contains two clades in which traps have evolved independently (Fig. 2, nodes 199 and 208), the common ancestor of the Pistia-clade apparently did not have a trap but already possessed papillate cells. This would imply that papillate cells were present before the emergence of slippery surfaces.

A trend similar to that in papillate cells can also be observed in elongated sterile flowers of the tribe Areae. In Sauromatum, sterile flowers situated below the staminate flowers act as osmophores (Hadacek and Weber, 2002). Moreover, we also found that sterile flowers of Typhonium produce scent and stain intensively after treatment with neutral red (Bröderbauer, unpublished data), which is used to detect osmophores (Vogel, 1963). In both taxa, sterile flowers are located within the floral chamber below the constriction of the spathe. By contrast, in Arum, the sterile flowers, which are present below and above of the staminate zone, are part of the trap (Knoll, 1926). They produce oil and are slippery, thereby preventing trapped insects from escape. Moreover, they act like a sieve that gives access to the spathe chamber only to insects of a certain size. Thus, sterile flowers apparently have shifted in function from osmophores to trapping devices in the Areae clade. The function of elongated sterile flowers in general varies in different clades. In Arisaema, sterile flowers present on the appendix help in the attraction of pollinators (Vogel and Martens, 2000), while in Bucephalandra they probably serve as protecting structures for the developing fruits (P. Boyce, Universiti Sains Malaysia, personal communication). In Dracontium and Amorphophallus the function of sterile flowers is unclear, but judging from their shape and position, a role in trapping insects seems unlikely in most species.

Movements of the spathe during or after anthesis are ubiquitous in Araceae (Mayo et al., 1997). In genera such as Dieffenbachia (Young, 1986) and Alocasia (Miyake and Yafuso, 2003), the constriction closes after the pollen release. These movements are thought to force the pollinators to leave the inflorescence and also to protect developing fruits (Mayo et al., 1997). The closure of the inflorescence during anthesis to imprison pollinators might result simply from a change in the timing of the spathe closure. In Cryptocoryne and Lagenandra, the spathe margins are connate and are not able to constrict. Instead, the seclusion of the chamber is achieved by the movement of a specialized extension of the spathe margin, the so-called flap (Ørgaard and Jacobsen, 1998). Besides their function in trapping, spathe movements can also be important for the release of pollinators. In Arisaema and Pinellia, insects are set free from the trap by spathe movements that result in the formation of a secondary opening (Vogel and Martens, 2000). This is necessary because in these traps slippery surfaces (i.e., epicuticular wax crystalloids) do not wither, thus preventing the insect’s escape through the still slippery entrance of the chamber.

Evolutionary history of functional types of traps

We found that inflorescence traps have evolved at least 10 times independently in the Araceae. Traps are not restricted to taxa with highly synorganized inflorescences but also occur in the subfamily Lasioideae. In this clade and in several other lineages, the spadix is not differentiated and bears bisexual flowers only. Moreover, the spathe only forms a rudimentary chamber without a constriction in most taxa of Lasioideae. Unisexual flowers appear in the Stylochaeton-clade and are prevalent in the Aroideae. Here, an increasing synorganization of spadix and spathe can be observed, with the pistillate flowers being enclosed in the spathe chamber, the sterile flowers leveling with the spathe constriction and the staminate flowers facing the spathe blade. Despite these morphological differences, convergent evolution has led to the formation of traps that function in a similar way in distinct clades. Perhaps the most astonishing examples for convergent evolution are the traps of the Zomicarpa type in Dracontioides (bisexual flowers), Zomicarpa, and Arisaema (unisexual flowers) (Vogel and Martens, 2000). A second trap type, which is present in bisexual (i.e., Lasioideae) and unisexual (i.e., Aroideae) clades, is the Arisarum type. This type represents an imperfect trap because insects glide down slippery surfaces and fall into the spathe chamber. They are, however, not imprisoned inside because they can escape by climbing the spadix (Vogel, 1978). The Arisarum type prevails in subfamily Lasioideae. We suppose that the evolution of perfect traps is less probable in this clade due to the lower degree of synorganization of spathe and spadix. Nevertheless, a transition from imperfect to perfect traps occurred within Lasioideae in Dracontioides desciscens. In contrast to the imperfect traps of the same clade, the spadix is completely hidden inside the spathe, and a secondary exit is formed by the opening of the convolute spathe margins. This example shows that imperfect traps may serve as a precursor for perfect traps. This tendency is supported by the number of transitions, which occurred most frequently from nontraps to traps of the imperfect Arisarum type and next most frequently from the Arisarum type to the Zomicarpa type.

The purpose of imperfect traps is to ensure that insects lured to an inflorescence will have contact with flowers before departing, leading to pollen transfer (Faegri and Van der Pijl, 1971). However, pollination success will be greatly improved in traps in which the insects are forced to stay inside the floral chamber, thus depositing cross pollen on the stigmas and removing pollen from the anthers more effectively (Lack and Diaz, 1991). Therefore, traits that ensure the retention of pollinators may be favored by selection in imperfect traps, facilitating the evolution of true traps. However, the presence of such a precursory imperfect stage could not be found in Stylochaeton and subfamily Aroideae. It remains uncertain whether it simply did not exist or it transitioned rapidly into a perfect trap. Nevertheless, we can still observe different degrees of synorganization. For example, Arum type traps are derived from the Typhonium type, in which the sterile flowers serve as osmophores not involved in trapping. The closure of the floral chamber is reached by a narrowing of the spathe constriction. In contrast, sterile flowers have become part of the trap in the Arum type, replacing the function of the spathe movements. Moreover, in the Arum type, the fertile part of the spadix is completely hidden within the spathe chamber, while at least in some taxa of the Typhonium type, staminate flowers are situated above the constriction of the spathe chamber.

Shifts from traps to nontraps are rare within Araceae. The only known example is found in the genus Arum, which mainly consists of deceptive traps pollinated by flies and beetles (Gibernau et al., 2004). Arum creticum, however, has shifted to bee pollination and rewards its visitors with pollen during the staminate phase of anthesis (Diaz and Kite, 2006). However, bees are still trapped during the pistillate phase to secure the transfer of outcross pollen onto the stigma. The absence of true transitions from traps to rewarding inflorescences indicates that trap pollination is an evolutionary stable condition within the Araceae.

Association between traps and pollinators

The ancestral pollinator type of Araceae could not be resolved unambiguously. Most clades originated from beetle- or fly-pollinated ancestors, with flies prevailing in the clades occurring after node 130 (Fig. 2), and beetles prevailing in the remaining clades. Bees serve as pollinators only in the subfamilies Monsteroideae and Pothoideae. The inflorescence traps in Araceae are known to be pollinated by beetles or flies and in some taxa by both occurring together. In most cases, these are saprophilous species (Gibernau, 2003). An obvious reason for the evolution of trap pollination is a change toward deceptive pollination, as insects will soon leave a flower when putative rewards are revealed to be a fake (Faegri and van der Pijl, 1971; Dafni, 1984). Chartier (2011) showed that, in Araceae, deceit pollination was derived from mating mutualism involving beetle pollination as well as from nursery mutualism involving flies, as was postulated by Stebbins (1970). But does pollination by a certain type of insect make a change to a deceptive trap more likely? We found that trap pollination in Araceae is correlated with pollination by flies rather than beetles. According to our ancestral state reconstructions, the common ancestors of clades with traps were pollinated by flies in the majority of cases. Interestingly, most changes from nontraps to traps were not associated with a simultaneous change in pollinator type but happened within fly-pollinated clades. Gibernau et al. (2010) showed that in several taxa with traps floral traits match those of mutualistic taxa pollinated by flies, indicating that trap pollination is embedded in the pollination syndrome of myophily. For example, traps of the Schismatoglottis type in Schismatoglottis and Colocasia are embedded in clades where nursery pollination involving flies prevails (Chartier, 2011). In contrast to other traps in Araceae, their pollinators (flies of the drosophilid genus Colocasiomyia) are not deceived. Their reward is a brood site (Toda and Okada, 1983; Takenaka et al., 2006; Toda and Lakim, 2011). The flies lay their eggs between the flowers, and larvae develop inside the decaying inflorescence. Contrary to the situation in other trap types, adult flies can move freely within the inflorescence during the pistillate phase. However, after some time, the spathe closes completely and thus imprisons the drosophilids. After pollen release, the spathe opens abruptly and the flies depart (Cleghorn, 1913; Boyce and Wong, 2007; Bröderbauer, unpublished manuscript). We conclude that in this case trapping is more important for ensuring efficient pollen export than for the pollen deposition on the stigma, which in any case is achieved by egg-laying flies.

A scenario with nursery mutualism as a precursor to trap pollination is also probable in other clades. Based on Chartier’s (2011) reconstruction of plant–pollinator interactions in Araceae we can infer that nursery mutualism was also present in the common ancestor of traps in the Arum clade. An example for trap pollination through deception of drosophilids in an extant member of Areae is found in Arum palaestinum (Stökl et al., 2010). However, deception of fruit flies in this species is probably derived from trap pollination by saprophilous flies (Linz et al., 2010). Nursery pollination by drosophilid flies is also found in Aristolochia (Sakai, 2002). While most species of Aristolochia form deceptive traps, Drosophila spp. pollinating A. maxima do not get retained but deposit their eggs in the flowers. These findings suggest that transitions between nursery mutualism and brood-site mimicry could be a common phenomenon. A shift to saprophilous pollinators can be achieved by simple changes in floral scent (Shuttleworth and Johnson, 2010). As floral odors are very diverse in the Araceae (Kite et al., 1998; Stökl et al., 2010; Schiestl and Dötterl, 2012) such changes in floral scent have probably occurred independently several times.

Further hypotheses that could explain our finding of a correlation between flies and trap pollination relate to the differential behavior of flies and beetles. Knoll (1926) and Bown (2000) argue that flies are much more agile and therefore have to be arrested to transfer pollen. In contrast, beetles tend to stay in flowers for longer intervals voluntarily (Dafni, 1984; Willmer, 2011). In addition, many chamber flowers/inflorescences offer solid food rewards for beetles (Proctor et al., 1996; Gibernau et al., 1999; Bernhardt, 2000), which cannot be consumed by flies.

Conclusions

The repeated emergence of morphological traits that facilitate trap pollination has led to the evolution of inflorescence traps at least 10 times, such that it is found in at least 27 genera of Araceae. On several occasions, the formation of trapping devices resulted from a shift of function in already existing inflorescence characters. Various functional types of traps evolved independently in different clades. In at least some of these clades, imperfect traps predated the evolution of perfect traps, and elaborate traps were derived from less complex ancestors. The evolution of traps is correlated with fly pollination. Nursery mutualism between aroid inflorescences and drosophilid flies is likely to be a precursor for the evolution of traps. Further studies on plant–pollinator interactions in such nursery mutualisms are needed to detect drivers for the evolution of floral traps in Araceae and elsewhere.

Appendix 1. List of plant material investigated under light and scanning electron microscope. Specimens stored in spirit collections are indicated by an asterisk.

Taxon; Voucher (Herbarium).

Aglaonema modestum Schott ex Engl.; 0064899 (WU)*. Aglaonema nebulosum N.E. Br.; 0064900 (WU)*. Alocasia acuminata Schott; 0064901 (WU)*. Alocasia lauterbachiana (Engl.) A. Hay; 0064902 (WU)*. Alocasia odora (Lindl.) K. Koch; 0064903 (WU)*. Alocasia portei Schott; Bogner 1768 (M). Ambrosina bassii L.; 0064905 (WU) *. Amorphophallus atrorubens Hett. & Sizemore; 0064906 (WU)*. Amorphophallus henryi N.E. Br.; 0064908 (WU)*. Amorphophallus konjac K. Koch; 0064910 (WU) *. Amorphophallus longituberosus (Engl.) Engl. & Gehrm.; 0064912 (WU)*. Amorphophallus mossambicensis (Schott ex Garcke) N.E. Br.; 0064913 (WU)*. Amorphophallus myosuroides Hett. & A. Galloway; 0064914 (WU)*. Amorphophallus palawanensis Bogner & Hett.; 0064917 (WU)*. Amorphophallus polyanthus Hett. & Sizemore; 0064918 (WU)*. Amorphophallus stuhlmannii (Engl.) Engl. & Gehrm.; 0064919 (WU)*. Amorphophallus taurostigma Ittenbach, Hett. & Bogner; 0064920 (WU)*. Amorphophallus variabilis Blume; 0064921 (WU)*. Amorphophallus yunnanensis Engl.; 0064922 (WU)*. Anadendrum affine Schott; 012384 (NCY)*. Anaphyllopsis americana (Engl.) A. Hay; Barabé et al. 258 (MT). Anchomanes dalzielii N.E. Br.; 0064924 (WU)*. Anchomanes difformis (Blume) Engl.; 012388 (NCY)*. Anchomanes giganteus Engl.; 012389 (NCY)*. Anthurium magnificum Engl.; 0064925 (WU)*. Anthurium nymphaeifolium K. Koch & C.D. Bouché; Bogner 762 (M). Anthurium pedatum (Kunth) Engl. ex Kunth; Bogner 2956 (M). Anubias gigantea A. Chev. ex Hutch.; 0064928 (WU)*. Anubias gilletii De Wild. & T. Durand; Bogner 108 (M). Arisaema fargesii Buchet; 0064931 (WU)*. Arisaema ghaticum (Sardesai, S.P. Gaikwad & S.R. Yadav) Punekar & Kumaran; 0064932 (WU)*. Arisarum proboscideum (L.) Savi; 0064933 (WU)*. Arisarum vulgare O. Targ. Tozz.; 0064934 (WU)*. Arophyton crassifolium (Buchet) Bogner; 0064935 (WU)*. Arophyton humbertii Bogner; 0064936 (WU)*. Arum cylindraceum Gasp. in G. Gussone; 0064941 (WU)*. Arum italicum Mill.; 0064947 (WU)*. Arum nigrum Schott; 0064949 (WU)*. Asterostigma lividum (Lodd.) Engl.; 0064951 (WU)*. Biarum carratracense (Willk.) Font Quer; 0064951 (WU)*. Biarum tenuifolium (L.) Schott in H.W. Schott & S.L. Endlicher; 0064954 (WU)*. Caladium bicolor (Aiton) Vent.; RMP 3137 (FRP). Caladium lindenii (André) Madison; Bogner 2338 (M). Caladium steudneriifolium Engl.; 0064958 (WU)*. Calla palustris L.; 0064959 (WU)*. Callopsis volkensii Engl.; 0064960 (WU)*. Carlephyton glaucophyllum Bogner; RMM 124 (FRP). Chlorospatha croatiana Grayum; 0064963 (WU)*. Colletogyne perrieri Buchet; 0064964 (WU)*. Colocasia affinis Schott; 0064966 (WU)*. Colocasia esculenta (L.) Schott in H.W. Schott & S.L. Endlicher; 0064967 (WU)*. Colocasia fallax Schott; 0064968 (WU)*. Colocasia fontanesii Schott; 0064969 (WU)*. Colocasia gigantea (Blume) Hook.f.; Bogner 427 (M). Culcasia saxatilis A. Chev.; Bogner 2727 (M). Cryptocoryne longicauda Becc. ex Engl.; 0064971 (WU)*. Cryptocoryne pontederiifolia Schott; Bogner 1739 (M)*. Cyrtosperma ferox N.E. Br. & Linden; Bogner 2131 (M). Cyrtosperma johnstonii (N.E. Br.) N.E. Br.; 1978.3.532 (NCY). Dieffenbachia bowmannii Carrière; 012504 (NCY)*. Dieffenbachia seguine (Jacq.) Schott in H.W. Schott & S.L. Endlicher; 012506 (NCY)*. Dieffenbachia oerstedii Schott; 0064978 (WU)*. Dracontioides desciscens (Schott) Engl.; 1994.3.770 (NCY). Dracontium amazonense G.H. Zhu & Croat; H.AR.83 (FRP). Dracontium asperum K. Koch; Bogner 2793 (M). Dracontium bogneri G.H. Zhu & Croat; 0064982 (WU)*. Dracontium nivosum (Lem.) G.H. Zhu in R.H.A. Govaerts & D.G. Frodin; 012516 (NCY)*. Dracontium polyphyllum L.; 0064984 (WU)*. Dracontium prancei G.H. Zhu & Croat; Bogner 1132 (M). Dracontium soconuscum Matuda; RMP 2233 (FRP). Dracontium spruceanum (Schott) G.H. Zhu; RMP 2162 (FRP). Dracunculus canariensis Kunth; 0064041 (WU)*. Dracunculus vulgaris Schott in H.W. Schott & S.L. Endlicher; 0064988 (WU)*. Filarum manserichense Nicolson; 0064990 (WU)*. Gonatopus boivinii (Decne.) Engl. in A.L.P. de Candolle & A.C.P. de Candolle; 0064991 (WU)*. Gorgonidium cf. intermedium (Bogner) E.G. Gonç.; 0064992 (WU)*. Hapaline cf. benthamiana Schott; 0064993 (WU)*. Helicodiceros muscivorus (L.f.) Engl. in A.L.P. de Candolle & A.C.P. de Candolle; 0064994 (WU)*. Homalomena picturata (Linden & André) Regel; 0064996 (WU)*. Homalomena wallisii Regel; 0064998 (WU)*. Incarum pavonii (Schott) E.G. Gonç.; 0064999 (WU)*. Lagenandra praetermissa de Wit; 0065000 (WU)*. Lasia spinosa (L.) Thwaites; 0065001 (WU)*. Lysichiton americanus Hultén & St. John; 0065002 (WU)*. Monstera adansonii Schott; 1981.3.587 (NCY). Monstera obliqua Miq.; 2003.3.214 (NCY). Nephthytis afzelii Schott; Bogner 2998 (M). Nephthytis hallaei (Bogner) Bogner; 012564 (NCY)*. Nephthytis sp. 0065004 (WU)*. Philodendron martianum Engl.; 0065006 (WU)*. Philodendron pedatum (Hook.) Kunth; 0065007 (WU)*. Philodendron sodiroi N.E. Br.; 0065008 (WU)*. Philodendron squamiferum Poepp. in E.F. Poeppig & S.L. Endlicher; Bogner 1958 (M). Pinellia cordata N.E. Br.; 0065011 (WU)*. Pinellia peltata C. Pei.; 0065012 (WU)*. Pinellia ternata (Thunb.) Makino; 0065013 (WU)*. Piptospatha ridleyi N.E. Br. ex Hook.f.; 012664 (NCY)*. Pistia stratiotes L.; 0065014 (WU)*. Pothos junghuhnii de Vriese in F.A.W. Miquel; Bogner 1550 (M). Pseudodracontium latifolium Serebryanyi; 0065014 (M). Pseudodracontium sp. 0065016 (WU)*. Pseudohydrosme gabunensis Engl.; 0065019 (WU)*. Pycnospatha palmata Gagnep.; 0065020 (WU)*. Rhaphidophora angustata Schott; Bogner 2989 (M). Rhaphidophora decursiva (Rox.) Schott; 0065022 (WU)*. Remusatia hookeriana Schott; 0065023 (WU)*. Remusatia pumila (D. Don) H. Li & A. Hay; 0065024 (WU)*. Remusatia vivipara (Roxb.) Schott in H.W. Schott & S.L. Endlicher; 0065025 (WU)*. Sauromatum venosum (Dryand. ex Aiton) Kunth; 0065026 (WU)*. Schismatoglottis calyptrata (Roxb.) Zoll. & Moritzi in H. Zollinger; 0065028 (WU)*. Schismatoglottis multiflora Ridl.; Bogner 1453 (M). Schismatoglottis subundulata (Zoll. ex Schott) Nicolson; 0065030 (WU)*. Scindapsus lucens Bogner & P.C. Boyce; 012699 (NCY)*. Spathicarpa hastifolia Hook.; Bogner 2546 (M). Spathiphyllum cannifolium (Dryand. ex Sims) Schott; 0065032 (WU)*. Spathiphyllum wallisii Regel; 0065033 (WU)*. Stenospermation popayanense Schott; Bogner 463 (M). Steudnera henryana Engl.; 0065035 (WU)*. Steudnera kerrii Gagnep.; 2000.3.441 (NCY). Stylochaeton bogneri Mayo; Bogner 216 (M). Stylochaeton cf. hypogaeus Lepr.; 0065038 (WU)*. Stylochaeton zenkeri Engl.; 0065039 (WU)*. Symplocarpus foetidus (L.) Salisb. ex W.P.C. Barton; 0065040 (WU)*. Synandrospadix vermitoxicus (Griseb.) Engl.; 0065040 (WU)*. Syngonium macrophyllum Engl.; 012708 (NCY)*. Syngonium podophyllum Schott; 012709 (NCY)*. Taccarum caudatum Rusby; 0065042 (WU)*. Typhonium blumei Nicolson & Sivad.; 0065043 (WU)*. Typhonium sp. nov. 0065047 (WU)*. Typhonium trilobatum (L.) Schott; 0065046 (WU)*. Typhonodorum lindleyanum Schott; 0065048 (WU)*. Ulearum sagittatum Engl.; 0065049 (WU)*. Urospatha grandis Schott; RMP 1306 (FRP). Urospatha sagittifolia (Rudge) Schott; Bogner 2770 (M). Urospatha tonduzii Engl.; Bogner 1115 (M)*. Xanthosoma cubense (Schott) Schott; 0065053 (WU)*. Xanthosoma mariae Bogner & E.G. Gonç.; 0065055 (WU)*. Zamioculcas zamiifolia (Lodd.) Engl.; 0065056 (WU)*. Zantedeschia aethiopica (L.) Spreng.; 0065057 (WU)*. Zantedeschia albomaculata (Hook.) Baill.; 012714 (NCY)*. Zomicarpa riedelianum Schott; Vogel 54 (WU).