INTRODUCTION
Epiphytic vascular plants comprise a major proportion of the tropical flora (Kreft et al. Reference KREFT, KÖSTER, KÜPER, NIEDER and BARTHLOTT2004, Krömer et al. Reference KRÖMER, KESSLER, GRADSTEIN and ACEBEY2005), contributing up to one-third of vascular plant species in wet tropical forests (Gentry & Dodson Reference GENTRY and DODSON1987). Among plant families, Orchidaceae have a high number of epiphytes, with approximately 20 000 species in 800 genera (Dressler Reference DRESSLER1993). Some hypotheses have been proposed to explain the higher local richness of epiphytic orchids in tropical plant communities, such as recurrent environmental constraints preventing competitive exclusion (e.g. bark defoliation and tree falls, Benzing Reference BENZING1981) and vertical niche diversification (Gentry & Dodson Reference GENTRY and DODSON1987). A hypothesis that has recently received much attention is that of host-tree specificity (Blick & Burns Reference BLICK and BURNS2009, Callaway et al. Reference CALLAWAY, REINHART, MOORE, MOORE and PENNINGS2002, Laube & Zotz Reference LAUBE and ZOTZ2006). Although the preference of epiphytes for host trees has been broadly reported in the literature (Laube & Zotz Reference LAUBE and ZOTZ2006), few studies have tried to link the interactions of entire epiphyte assemblages with entire host-tree assemblages in a quantitative manner, searching for a general pattern at the community level (Blick & Burns Reference BLICK and BURNS2009, Burns Reference BURNS2007).
Species interactions may be depicted as networks (e.g. mutualistic, Bascompte et al. Reference BASCOMPTE, JORDANO, MELIÁN and OLESEN2003; commensalistic, Burns Reference BURNS2007; parasitic, Vázquez et al. Reference VÁZQUEZ, POULIN, KRASNOV and SHENBROT2005). In interaction networks, each species is connected to one or more other species in binary interaction matrices, in which one group of species is listed as rows and the other group is listed as columns (Proulx et al. Reference PROULX, PROMISLOW and PHILLIPS2005). Interacting pairs of species receive a ‘1’ in their corresponding cell, whereas non-interacting pairs of species receive a ‘0’ in the matrix (Proulx et al. Reference PROULX, PROMISLOW and PHILLIPS2005). Recently, general patterns in networks are emerging from many types of species interaction (Bascompte & Jordano Reference BASCOMPTE and JORDANO2007). For example, networks of mutualistic interactions have nested patterns (Bascompte et al. Reference BASCOMPTE, JORDANO, MELIÁN and OLESEN2003), whereas networks of antagonistic interactions generally have non-nested ones (Guimarães et al. Reference GUIMARÃES, RICO-GRAY, REIS and THOMPSON2006). Thus, networks of species interactions are a unifying framework to compare patterns of specialization within and among different types of ecological interaction (Bascompte & Jordano Reference BASCOMPTE and JORDANO2007, Proulx et al. Reference PROULX, PROMISLOW and PHILLIPS2005).
Nestedness is a specific type of asymmetric interaction among species. This pattern is characterized by: (1) species with many interactions forming a core of interacting species; (2) species with few interactions interacting only with species with many interactions; and (3) a lack of interactions among species with few interactions (Bascompte et al. Reference BASCOMPTE, JORDANO, MELIÁN and OLESEN2003, Guimarães et al. Reference GUIMARÃES, RICO-GRAY, REIS and THOMPSON2006). The nested pattern observed in networks can be the result of ecological and stochastic factors. In mutualistic interactions, nestedness is attributed to the relatively stable set of resources provided by the core of generalist-generalist associations, which allow a larger number of specialized interactions to persist (Ollerton et al. Reference OLLERTON, MCCOLLIN, FAUTIN and ALLEN2007). However, a great proportion of this nestedness may be explained simply by differences in relative abundance of the interacting species (Krishna et al. Reference KRISHNA, GUIMARÃES, JORDANO and BASCOMPTE2008, Vázquez et al. Reference VÁZQUEZ, MELIÁN, WILLIAMS, BLÜTHGEN, KRASNOV and POULIN2007). Recently, Burns (Reference BURNS2007) also found a nested pattern in a small epiphyte-host tree network. He suggested that a sequential colonization of epiphytic species that are less tolerant of moisture and nutrient stress might produce nestedness.
We studied 40 networks of epiphytic orchid species and host tree species in Brazilian gallery forests. Our general goal is to understand how epiphytic networks are structured. Among epiphytes, Orchidaceae are generally the most abundant family in tropical forests (Gentry & Dodson Reference GENTRY and DODSON1987). Despite the richness of this group, few studies have assessed the pattern of interaction between epiphytic orchid species and host tree species in tropical forests (Migenis & Ackerman Reference MIGENIS and ACKERMAN1993). We investigated whether the epiphytic orchid network is phylogenetically structured (i.e. whether there is a phylogenetic signal, sensu Blomberg & Garland Reference BLOMBERG and GARLAND2002). In general, plants show a high degree of evolutionary stasis (Qian & Ricklefs Reference QIAN and RICKLEFS2004) and niche conservatism (Ackerly Reference ACKERLY2003, Prinzing et al. Reference PRINZING, DURKA, KLOTZ and BRANDL2001). This is because recent common ancestry constrains the adaptive radiation of a lineage of organisms in several ways: (1) closely related taxa have had less time to evolve differences than distantly related taxa (phylogenetic inertia; Felsenstein Reference FELSENSTEIN1985); (2) developmental, physiological or architectural patterns may bias the kinds of traits that can evolve (structural constraints; Wake & Larson Reference WAKE and LARSON1987); and (3) low fitness of transitional forms between discrete adaptive optima may restrict the diversity of traits that evolve (adaptive constraints; Wright Reference WRIGHT1982). Thus, whether or not there is a phylogenetic signal in the number and in the similarity of interactions, closely related orchids are expected to share similar host-tree assemblages and closely related host trees are expected to house similar orchid assemblages.
To test this hypothesis, we tried to answer the following questions: (1) Is nestedness a characteristic of orchid–host tree networks? (2) Is the phylogenetic distance among orchid species correlated with the similarity of their host tree species? (3) Is the phylogenetic distance among host tree species correlated with the similarity of their orchid species? (4) Is the number of host tree species colonized by orchid species a trait conserved on the phylogeny of orchids? (5) Is the number of orchid species hosted by tree species a trait conserved on the phylogeny of trees?
METHODS
Ecological data
We surveyed 40 gallery forests in 23 municipalities of central São Paulo state, south-eastern Brazil (21°14′–22°32′S, 47°35′–48°32′W, Figure 1). The areas of the surveyed sites range from 1 to 200 ha. The altitude ranges from 550 m to 1000 m approximately. The regional climate is humid subtropical with alternating wet and dry seasons (Cwa, Köppen Reference KÖPPEN1948).
Figure 1. Location of the study site, in São Paulo state, Brazil (21°14′–22°32′S, 47°35′–48°32′W). The lines in the map delimit the territory of the municipalities. ● = municipalities of the study, × = surveyed areas.
In general, the gallery forests had a continuous canopy cover approximately 10 m tall, composed mostly of individuals of Calophyllum brasiliense, Magnolia ovata, Tapirira guianensis and Xylopia emarginata. Most of the host tree species were dominant in the middle layer of the gallery forests. The understorey was composed of a mix of sparsely distributed herbs, ferns and shrubs.
Between January 2006 and April 2008, we inspected with binoculars all trees along trails traversing the entire gallery forest sites. We choose this technique because the trail generally affords better visual access into tree crowns. Moreover, the accuracy of ground-based surveys in comparison with surveys from canopy walkways of epiphytes is often high (see Blick & Burns Reference BLICK and BURNS2009 and references therein). The duration of the inspections ranged from several hours (in the smallest sites) to 4 d (in the largest one). All observed individuals of orchids and host trees were collected with a canopy crane. The orchid specimens collected in the vegetative stage were cultivated in a greenhouse with mineral fertilization until they flowered. The flowering orchids were identified using important taxonomic literature about Brazilian Orchidaceae (Cogniaux Reference COGNIAUX, Martius, Eichler and Urban1893–1896, Hoehne Reference HOEHNE and Hoehne1940) and by comparing them with specimens in herbaria. The voucher samples of the orchid species were deposited at the herbaria of Universidade Federal de São Carlos (HUFSCar), Universidade Estadual de Campinas (UEC), and Universidade de São Paulo at Ribeirão Preto (SPFR). The host tree species were identified by comparing with specimens lodged in the herbaria of HUFSCar and UEC and also by using the literature.
Nestedness
Interspecific interactions can be described as networks in which species are nodes and interactions between any species pair are depicted as links (Bascompte et al. Reference BASCOMPTE, JORDANO, MELIÁN and OLESEN2003). We defined an epiphytic orchid network by a matrix describing interactions between orchid species in rows and host tree species in columns. An element a ij of such a matrix was ‘1’ if orchid species i grows upon tree species j, and ‘0’ otherwise. We constructed a matrix for each surveyed site. We included in the matrix the orchid species that were seen in more than two individuals of a certain host tree species. We assessed the nestedness of each matrix with the NODF index (Almeida-Neto et al. Reference ALMEIDA-NETO, GUIMARÃES, GUIMARÃES, LOYOLA and ULRICH2008). NODF is an acronym for nestedness metric based on overlap and decreasing fill and its value increases with nestedness (0 to 1, Almeida-Neto et al. Reference ALMEIDA-NETO, GUIMARÃES, GUIMARÃES, LOYOLA and ULRICH2008). The NODF has shown better statistical properties than traditional metrics (Almeida-Neto et al. Reference ALMEIDA-NETO, GUIMARÃES, GUIMARÃES, LOYOLA and ULRICH2008), such as T, the matrix temperature measure (Atmar & Patterson Reference ATMAR and PATTERSON1993).
We used two null models to assess the significance of the degree of nestedness we found for each site. In the first null model, the presences were randomly assigned to any cell within the matrix. As a consequence, we tested if the observed NODF was higher than expected for random networks with a similar number of interactions. In the second null model, the probability (C) that an orchid species i occurs on a tree species j depended on the observed number of interactions of both species, such that:
in which ki is the number of presences in row i, kj is the number of presences in column j, F is the number of columns and L is the number of rows. Therefore, we tested whether the observed NODF was higher than expected for random networks with similar heterogeneity of interactions among species. We computed the NODF and conducted the statistical analysis with the ANINHADO 3.0 program (Guimarães & Guimarães Reference GUIMARÃES and GUIMARÃES2006). The null-model approach of the ANINHADO was based on the comparison of real data with an empirical distribution obtained from the analysis of many algorithm-generated matrices (Guimarães & Guimarães Reference GUIMARÃES and GUIMARÃES2006).
Phylogenetic analysis
We included in the phylogenetic analysis the orchid and host tree species that were seen at more than two sampled sites. Consequently, we excluded from the analysis 20 orchid and 13 host-tree species. We constructed the phylogenetic tree for the orchid species based on the phylogenetic relationships of subtribes summarized by Chase et al. (Reference CHASE, BARRETT, CAMERON, FREUDENSTEIN, Dixon, Kell, Barrett and Cribb2003) from several published DNA studies. We established the branch lengths of the family based on minimum age from fossil data (40 million y, Wikström et al. Reference WIKSTRÖM, SAVOLAINEN and CHASE2001) and the branch lengths of subtribes, by evenly spacing the nodes of the constructed tree. Then we established the branch lengths of the orchid species by evenly dividing the branch lengths of the subtribes (Figure 2).
Figure 2. Phylogenetic tree assembled for the orchid genera sampled in gallery forests, southeastern Brazil. The phylogenetic relationships of subtribes followed the classification proposed by Chase et al. (Reference CHASE, BARRETT, CAMERON, FREUDENSTEIN, Dixon, Kell, Barrett and Cribb2003).
To test whether the closely related orchids shared similar host-tree assemblages, we first calculated the phylogenetic distances between all pairs of orchid species. We computed the distance from the estimated intervening branch lengths (in million y) of the phylogenetic tree of orchids with the PHYDIST module of the Phylocom software package (http://www.phylodiversity.net/phylocom/). We also computed the similarity (Jaccard index) of host tree species between all pairs of orchid species. Then we compared the correlation coefficient between pairwise phylogenetic distances and similarity of host tree species to a null model, in which the phylogenetic distances among orchid species were randomized (Mantel test, Manly Reference MANLY2004). The Mantel test of association was conducted in pairwise species similarity matrices, with 1000 randomizations.
We constructed the phylogenetic tree for the host tree species sampled with the Phylomatic software, a phylogenetic database and toolkit for the assembly of phylogenetic trees (Webb & Donoghue Reference WEBB and DONOGHUE2005). The tree generated was based on information from several published molecular phylogenies (Figure 3, Phylomatic reference tree R20050610). We assigned branch lengths to the phylogenetic tree using the BLADJ (Branch Length Adjustment) averaging algorithm of the Phylocom software package (http://www.phylodiversity.net/phylocom/). The branch length was based on minimum ages of nodes determined for genera, families and higher orders from fossil data (Wikström et al. Reference WIKSTRÖM, SAVOLAINEN and CHASE2001). The ages of the undated nodes were obtained by spacing the undated nodes evenly between dated nodes in the phylogenetic tree (Figure 3).
Figure 3. Summary of the phylogenetic tree assembled for the host tree species in gallery forests, southeastern Brazil. The relationship among families was based on Phylomatic reference tree R20050610 (Webb & Donoghue Reference WEBB and DONOGHUE2005).
To test whether the closely related host trees shared similar orchid assemblages, we first calculated the phylogenetic distances between all pairs of host tree species. We also computed this distance (in million y) with the module PHYDIST of the Phylocom software package (http://www.phylodiversity.net/phylocom/). We computed the similarity (Jaccard index) of orchid species between all pairs of host tree species. Then we compared the correlation coefficient between pairwise phylogenetic distances and similarity of orchid species to a null model, in which the phylogenetic distances among host tree species were randomized 1000 times (Mantel test, Manly Reference MANLY2004).
To test whether the number of interactions is a trait conserved on the phylogeny of the orchid and host tree species, we also used the mean number of host tree and orchid species per site for each orchid and host tree species respectively. Then we applied a test based on the variance of phylogenetic independent contrasts to assess whether the mean number of interactions tended to be phylogenetically conserved (i.e. phylogenetic signal) on the phylogeny of the orchid and host tree species (Blomberg et al. Reference BLOMBERG, GARLAND and IVES2003). If related species are similar to each other, the magnitude of the independent contrasts will generally be similar across the tree, resulting in a small variance of the contrast values (Blomberg et al. Reference BLOMBERG, GARLAND and IVES2003). The observed contrast variances were compared to the expectations under a null model of randomly swapping trait values across the tips of the tree. For a detailed description of comparative analyses using phylogenetically independent contrast, see Garland et al. (Reference GARLAND, HARVEY and IVES1992). We did this analysis with the Picante package (http://picante.r-forge.r-project.org) for the R environment (http://www.R-project.org). We used the number of interactions normalized by the standard deviation.
RESULTS
We sampled 105 orchid species and 132 host tree species (Appendixes 1, 2). The mean number of interactions per site for the orchid species was 8.7 ± 6.9 and for the host tree species was 6.4 ± 4.8. We found a general nested structure in the 40 networks of epiphytic orchids and host trees (Figure 4, mean NODF = 0.413 ± 0.080, mean P < 0.001 for both null models).
Figure 4. Network of interactions between epiphytic orchid species (rows) and host tree species (columns) in gallery forests, south-eastern Brazil. Rows and columns are arranged to maximize nestedness according to Atmar & Patterson (Reference ATMAR and PATTERSON1993).
Closely related orchid species did not share similar host tree assemblages. We did not observe significant correlation levels in the comparison between pairwise phylogenetic distances of orchids and similarity of host trees (r = −0.028, P = 0.363). Likewise, closely related host tree species did not house similar orchid assemblages. We did not observe significant correlation levels in the comparison between phylogenetic distances of host trees and similarity of orchids (r = 0.001, P = 0.323).
The mean number of interactions may not be a trait conserved in the phylogeny of orchids and of host trees. The change in the number of species interactions across the phylogenetic trees was not different from random (P = 0.149 for orchids and P = 0.359 for host trees). The values of the variance and of the mean of the random variances of the phylogenetic independent contrasts were, respectively, 0.027 and 0.032 for orchids and 0.140 and 0.172 for host trees.
DISCUSSION
The nested pattern shows that the host specificity of orchids is small and most of the interactions occur among generalist orchid species and generalist host tree species. While the concept of species-specificity can thus be rejected, the extreme alternative – that the interacting orchids and host trees are not a random subset of the regional species pool – can be dismissed as well. The nestedness was higher than expected by chance, suggesting that the interactions among epiphytic orchid and host tree species are deterministically assembled and some process may be involved in the structuring of epiphytic networks. However, although the plants present commonly a high degree of niche conservatism (Prinzing et al. Reference PRINZING, DURKA, KLOTZ and BRANDL2001), the networks of epiphytic orchids and host trees did not seem to be phylogenetically structured.
Nestedness seems to be an overall characteristic of epiphytic networks. Burns (Reference BURNS2007) also demonstrated that the epiphytic networks of New Zealand forests had a nested pattern of interactions. However, other plant–plant interactions may not show this pattern. Mistletoes and lianas, for example, presented fewer interactions than expected by chance in New Zealand forests, indicating specialized host preferences (Blick & Burns Reference BLICK and BURNS2009, but see Arruda et al. Reference ARRUDA, CARVALHO and DEL-CLARO2006). The nestedness hypothesis states that the core of interacting generalist species provides a stable set of resources for the inclusion of more specialized species (Ollerton et al. Reference OLLERTON, MCCOLLIN, FAUTIN and ALLEN2007). This hypothesis explains mutualistic interactions, in which nestedness describes a highly diffuse coevolution that may diminish the risk of extinction of specialist species (Bascompte et al. Reference BASCOMPTE, JORDANO, MELIÁN and OLESEN2003). In epiphytic networks, however, interactions among species are not mutually advantageous (Thompson Reference THOMPSON2005). Consequently, the nestedness in epiphytic networks would not diminish the risk of extinction of specialist species and a strong nested pattern may not emerge. The low NODF values we found support this assertion.
We did not find a phylogenetic signal in the epiphytic orchid–host tree network. Again, one plausible explanation for this result is that interactions among species in commensalistic networks are not mutually advantageous. As a consequence, natural selection acting on commensalisms may not favour the convergence and complementarity of traits in interacting species, such as in mutualisms (Thompson Reference THOMPSON2005). On the contrary, natural selection acting on epiphyte-host tree networks tends to favour convergence of traits only among epiphytes (Thompson Reference THOMPSON2005). A phylogenetic signal was found in 40% of mutualistic networks (Rezende et al. Reference REZENDE, JORDANO and BASCOMPTE2007). In general, mutualistic networks with a higher number of species showed frequent phylogenetic signals (Rezende et al. Reference REZENDE, JORDANO and BASCOMPTE2007). In this study, we also analysed a rich epiphytic orchid–host tree network. Thus, phylogenetic signals may be rare in commensalistic networks and other factors can be expected to determine the nestedness.
Burns (Reference BURNS2007) suggested that the sequential colonization of epiphytic species may produce nestedness. He argued that some species are often the first to colonize host trees and that these pioneering species favour the colonization of other epiphytes. Although purely speculative, similar successional processes may occur in epiphytic orchid assemblages. Orchid species are highly dependent upon mycorrhizal fungi to provide the resources necessary for germination and seedling growth (Rasmussen Reference RASMUSSEN2002, Smith & Read Reference SMITH and READ1997). Most orchid mycorrhizal fungi are saprophytic (Roberts Reference ROBERTS1999) and orchid seed germination depends on the presence of these fungi in the substrate (Leake Reference LEAKE1994). As mycorrhizal fungi are associated with the roots, the establishment of an orchid species on a host tree may facilitate the establishment of other species, by increasing the chance of mycorrhizal colonization. The hyphae of these fungi may cover the bark of host tree branches (Smith & Read Reference SMITH and READ1997) and favour the establishment of other orchids. Successional orchid progressions have been documented by examining orchid epiphytes on branches of different age classes. Generally, younger host trees and branches support different orchid associations than older substrates (Catling & Lefkovitch Reference CATLING and LEFKOVITCH1989, Catling et al. Reference CATLING, BROWNELL and LEFKOVITCH1986). We also observed that some orchid species are often pioneers in the colonization of host trees (e.g. Ionopsis utricularioides, Polystachya spp. and Rodriguesia decora, Appendix 1). Thus, succession may play an important role in the assembling of plant–plant commensalistic networks.
In general, epiphytes occurring on older branches require the mantle of dead organic matter that accumulates on older bark (Johansson Reference JOHANSSON1974). Consequently, host trees with large branches tend to house more epiphyte species (Johansson Reference JOHANSSON1974, Migenis & Ackerman Reference MIGENIS and ACKERMAN1993). Ingram & Nadkarni (Reference INGRAM and NADKARNI1993) found a positive correlation between branch circumference and epiphytic organic matter in a tropical forest. Thus, in epiphytic networks, the nestedness may also be a consequence of the thickness distribution of the host trees in tropical forests. The host tree species that had more orchid species (e.g. Calophyllum brasiliense and Magnolia ovata) are often the species with higher basal area in Brazilian gallery forests (Guarino & Walter Reference GUARINO and WALTER2005, Marques et al. Reference MARQUES, SILVA and SALINO2003). Moreover, there is also evidence that differences in abundance among species might generate nestedness (Vázquez et al. Reference VÁZQUEZ, POULIN, KRASNOV and SHENBROT2005). The host tree species with more orchid species are also frequently dominant species in Brazilian gallery forests (Guarino & Walter Reference GUARINO and WALTER2005, Marques et al. Reference MARQUES, SILVA and SALINO2003). Therefore, part of the nestedness of orchid–host tree networks may be also due to differences in abundance of host tree species.
Host tree characteristics influence epiphyte establishment and survival and, consequently, the distribution of epiphytes on trees (see López-Villalobos et al. Reference LÓPEZ-VILLALOBOS, FLORES-PALACIOS and ORTIZ-PULIDO2008 for references). Thick bark tends to favour the establishment of epiphytes (López-Villalobos et al. Reference LÓPEZ-VILLALOBOS, FLORES-PALACIOS and ORTIZ-PULIDO2008). Thus, the functional attributes of the host tree species might also have contributed to the nestedness. We could not quantify empirically these functional traits in sampled host tree species. However, some trees with many orchids have thick bark (e.g. Calophyllum brasiliense, Inga edulis, Magnolia ovata and Xylopia emarginata, Appendix 2) which might favour the establishment of a larger number of species.
Probably because of logistical difficulties related to studying epiphytes, ecological research on epiphytic interactions often focuses on population- or species-level processes. There is still a complete lack of information on the biology of most epiphytes, so that we can only speculate on the reasons of the nestedness in epiphytic networks. However, recent research has begun to explore community-level processes in epiphytes (Blick & Burns Reference BLICK and BURNS2009, Burns Reference BURNS2007). The results reported here support that the phylogenetic history of species does not generate the nestedness in orchid–host tree networks. Further studies should test the hypotheses we presented.
ACKNOWLEDGEMENTS
We are grateful to the National Research Council (CNPq), for the scholarship granted to the first author and the Office to Improve University Research (CAPES) for the scholarship granted to the second author; to the Brazilian environmental agency Ibama and local farmers, for research permission; and to A. Medeiros, A. Alves and C. A. Casali, for helping in the field work.
Appendix 1. Orchid species sampled and mean number of host tree species in which they occurred in gallery forests, south-eastern Brazil.
Appendix 2. Host tree species sampled and mean number of epiphytic orchid species observed in gallery forests, south-eastern Brazil.
