By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
*Author to whom correspondence should be addressed. K. C. Burns. Tel. +64 4 463 5339. Fax +64 4 463 5331. E-mail firstname.lastname@example.org.
1Trophic relationships are often depicted as networks, in which species are connected by links representing predatory interactions. Network analyses have also been applied to non-trophic species interactions, such as pollination and seed dispersal mutualisms, to summarize community-level patterns in mutually beneficial interactions. However, the processes responsible for common network properties such as degree distributions (the number of interactions maintained by each species in the network) and nestedness (the tendency for specialists to interact with perfect subsets of the species interacting with generalists) are poorly understood.
2I evaluated the network properties of a novel type of species interaction, commensal interactions between epiphytes and their host trees. I quantified the occurrence of 19 vascular epiphyte species on seven host tree species in a metacommunity of epiphytes in New Zealand. I then developed null models to test whether observed network properties result from the frequency of species interactions, which is a commonly cited but rarely tested explanation for degree distributions and nestedness.
3Results showed that degree distributions of both epiphyte and host tree species were consistent with a null model that randomly populated hosts with epiphytes. Several null models that test for nestedness by randomly linking pairs of interacting species showed significantly higher nestedness in the real community. In fact, the observed level of nestedness was among the highest yet recorded for any type of ecological interaction. A second null model that randomized species interactions according to their overall frequencies of occurrence in the community confirmed support for nestedness.
4Results indicate that the degree distributions were consistent with randomized expectations; the number of interactions maintained by both epiphyte and host tree species appears to be determined by their frequency of occurrence in the community. However, interaction frequencies could be ruled-out as an explanation for nestedness. Nested epiphyte–host interactions may instead result from a predictable sequence of epiphyte succession, wherein earlier colonists ameliorate environmental conditions within host trees for later recruiting species.
Interactions between predators and prey (Pimm 1982), pathogens and hosts (Vázquez et al. 2005), and mutualists (Bascompte et al. 2003) are often depicted as networks, which summarize community-level relationships between interacting species. Viewing species interactions as networks provides a unifying framework to compare patterns in species interactions within and among different types of ecological interactions (Proulx et al. 2005; Lewinsohn et al. 2006). Here, I extend network analyses to cover a previously univestigated type of interaction, commensalistic plant-plant relationships between epiphytes and their host trees.
Interaction networks can be characterized by two general properties, the number of links (species interactions) each species maintains, and the composition of species united by links (also see Newman 2003; Boccaletti et al. 2006). Community-level patterns in the number of links maintained by each species are commonly referred to as ‘degree’ distributions (see Proulx et al. 2005). Degree distributions may be structured by a variety of deterministic processes, such as phenotypic mismatching between mutualist species that results in ‘forbidden’ links (Jordano et al. 2003; Stang et al. 2006, 2007). An alternative explanation is that degree distributions result from differences in the frequency of species interactions (Vázquez & Aizen 2003; Vázquez et al. 2005). Nearly all ecological communities are assembled from a small number of common species and a large number of rare species (see Hubbell 2001). Assuming there are no constraints on species interactions, such as deterministic processes that generate forbidden links, more abundant species will form greater numbers of links than less abundant species (Vázquez 2005).
Patterns in the second type of network property, the composition of species united by links, are usually evaluated by testing for ‘nestedness’ (Bascompte et al. 2003; Thompson 2005). The concept of nestedness was originally developed in the context of island biogeography (Darlington 1957; Atmar & Patterson 1993; Lomolino 1996), where it is used to describe archipelagos where the species inhabiting depauperate islands form perfect subsets of the species inhabiting diverse islands (see Burns 2007). When the concept is applied to species interactions, nestedness occurs when the partners of species with smaller numbers of links form perfect subsets of those with larger numbers of links.
As with degree distributions, the processes responsible for nestedness are poorly resolved. Nestedness in mutualistic networks could result from deterministic processes such as the ‘nestedness hypothesis’ (sensu Ollerton et al. 2007), which argues that ecological relationships between species with larger numbers of links (generalists) increases the availability of resources for species with fewer links (specialists), thereby allowing them to persist (Bascompte et al. 2003; Thompson 2005). Nestedness could also result from differences in the frequency of species interactions (Andrén 1994; Fischer & Lindenmayer 2002). Randomly simulated island communities comprised of differentially abundant species are often nested (Higgins et al. 2006). Therefore, if there are no constraints on the species composition of interacting species, nestedness could be generated from the frequency of interactions between species in the community.
Degree distributions and nestedness characterize distinct aspects of interaction networks. Although analyses of each characteristic may yield similar conclusions, null model analyses of degree distributions could differ substantially from null model tests for nestedness. A single matrix containing degree distributions that are consistent with a frequency of interaction null model might nevertheless show non-random variation in the composition of species united by those links.
I quantified the distribution of epiphytes in a New Zealand forest to evaluate the network properties of epiphyte–host interactions. Epiphyte communities on individual trees within forest stands can be usefully conceptualized as metacommunities, or groups of spatially isolated communities connected by dispersal (Wilson 1992; Leibold et al. 2004). To quantitatively characterize the epiphyte metacommunity, I inspected 274 trees belonging to seven common host species for the presence of 19 vascular epiphyte species. Observed degree distributions and levels of nestedness were then compared with null models that randomly populated host trees with epiphytes to determine whether the frequency of species interactions can explain observed network properties.
study site and species
All data were collected on the southern tip of the North Island, New Zealand, in Otari-Wilton's Bush (41°14′ S, 174°45′ E), which contains 75 hectares of mature and regenerating conifer-broadleaf forest. Detailed descriptions of the site's climate, history, topography and vegetation can be found elsewhere (Dawson 1988; Wardle 1991; Burns & Dawson 2005).
The forest is dominated by five canopy tree species and two canopy emergent trees (see Appendix 1 in Supplementary Material). Tree ferns, Dicksonia spp. (Dicksoniaceae) and Cyathea spp. (Cyatheaceae) also provide substrate for epiphytes (see Moran et al. 2003), but they were not considered because of strong differences in their growth habits and trunk morphology. There are 12 common epiphyte species at the site (see Appendix S1). They consist of four fern species, three orchid species, two angiosperm shrubs and a large species of clubmoss. There are also two species of lily that often grow into massive ‘nests’ that contain large quantities of humic soil (Wardle et al. 2003). A variety of trees, shrubs, ferns and sedges that commonly occur on the forest floor occasionally grow epiphytically.
A number of smaller epiphytes including mosses, liverworts, lichens and filmy ferns (Hymenophyllaceae) form integral components of New Zealand epiphyte communities (Dickinson et al. 1993; Hofstede et al. 2001). However, they were omitted from consideration because their small size makes them difficult to sample accurately without intensive effort on small spatial scales. Large vascular plants are easier to locate from farther away, making them more amenable to community-level analyses.
The study of epiphytes is fraught with logistical difficulties. Complete inventories of host trees require specialized equipment and intensive sampling effort (see Lowman 2001). Perhaps as a result, community-level studies of epiphytes are rare and network properties of epiphyte metacommunities are unknown. In an attempt to overcome these logistical hurdles, I used several sampling techniques to inventory epiphyte–host species interactions. First, I conducted ground-based searches using binoculars to document the occurrence of epiphyte species. All trees growing within 5 m on either side of a 0.5 km trail traversing the reserve were sampled (see Burns & Dawson 2005 for a detailed description of methodology). Smaller trees (< 10 cm diameter at breast (1.4 m) height) usually lacked epiphytes and were not considered. Host species that were found five or fewer times were also omitted because an accurate representation of the epiphyte species associated with them could not be obtained due to their rarity.
To standardize census effort among host trees, the time taken to inventory each tree was recorded. When it seemed that all epiphyte species inhabiting each tree had been recorded, a further search was made for one-third of the time already spent searching. If a new species was encountered during this time, an additional one-third of the total survey time was again spent searching in an attempt to ensure an adequate, consistent survey of each tree.
Next, I conducted complete inventories of several host trees from a canopy-walkway to identify the accuracy of ground-based surveys. The crowns of 22 trees were accessible from the walkway, which afforded an unimpeded view of three species, Beilschmiedia tawa (n = 10), Elaeocarpus dentatus (n = 6) and Knightia excelsa (n = 6). Comparisons between ground-based and canopy walkway surveys showed that over 90% of epiphyte occurrences were located from the ground. No consistent biases were observed among the three host tree species and no epiphyte species was particularly prone to being overlooked (see Burns & Dawson 2005). However, epiphytes were more likely to be missed in large trees, indicating that ground-based surveys of large trees are likely to underestimate their true epiphyte diversity. Most epiphytes that were not observed from the ground were small seedlings growing on the upper surfaces of branches at the top of tree crowns. Consequently, comparisons with the canopy walkway showed that while ground-based surveys underestimated epiphyte diversity on larger trees, they succeeded in identifying most epiphyte occurrences and were not biased against particular species of epiphytes.
Under this sampling regime, accurate comparisons of completely inventoried trees are impossible. However, documenting the network properties of epiphyte–host interactions requires complete surveys of species interactions, not complete inventories of individual trees. To evaluate the completeness of species-based surveys, I conducted rarefaction analyses (see Gotelli & Colwell 2001) using RAREFACT 1.0. This technique randomly selects subsamples of individual epiphyte occurrences (i.e. the presence of an epiphyte species on an individual host tree) from the total pool observed. It then calculates the number of species interactions (i.e. links) per subsample of occurrences on individual trees. This process is repeated across a range of occurrence values to obtain a rarefaction ‘curve’, which relates the accumulation of links to the accumulation of epiphyte occurrences.
I used a hyperbolic function (y = ax/(b + x), where a and b are fitted constants) to characterize the relationship between the number of links and the number of individual occurrences observed, because it consistently provided an accurate fit to the data. Curves were fit to rarefaction samples for each host species, and the community as a whole, using SigmaPlot (2002). The resulting regression equation was then used to characterize sampling efficiency by extrapolating the number of additional epiphyte occurrences required to obtain an additional link, which I expressed as a percentage of the total number of epiphyte occurrences observed during field sampling.
I tested for non-random patterns in degree distributions by deriving a null model that randomly placed epiphytes onto host trees. The null model involved computing null interaction matrices (R) by randomly assigning each individual epiphyte occurrence to a particular host tree species, and then comparing the resulting null matrix with the observed matrix. To obtain null interaction matrices, all individual epiphyte occurrences (O) from epiphyte species (i) were randomly assigned to a host tree species (j) according to the probability (P), which was equated to the fraction of all individual epiphyte occurrences maintained by that host species:
Therefore, if a particular host species housed 50% of all individual epiphyte occurrences, it would have a 50% chance of being colonized by each simulated epiphyte occurrence, and the probability that a link would be formed between each epiphyte-host species pair was proportional to their overall frequency of interaction. This procedure results in a null interaction matrix with cell values representing the number of times (i.e. trees) each species pair was randomly placed together. To convert each null interaction matrix (R) into binary matrix (Rβ) each cell of the null matrix (Rij) was set equal to ‘1’ if its corresponding species pair interacted one or more times and ‘0’ otherwise:
This procedure was iterated 500 times in Mathematica (Wolfram 1999) and the average number of links formed by each epiphyte and host tree species was taken as their expected degree value.
Congruency between observed and expected numbers of links was assessed with least-squares regression in SPSS (2002); 95% confidence limits for estimates of the slope and y-intercept of relationships between observed and expected degree values were calculated to evaluate whether expected values consistently predicted observed values across the full range of expected values. A one-to-one relationship between variables would suggest that deterministic processes such as forbidden links are unimportant in determining degree distributions and that the number of links maintained by each player in the interaction results from random interactions between differentially abundant species.
I tested for nestedness in epiphyte–host interactions using several different null models. First, I searched for pattern in the observed binary interaction matrix using Leibold & Mikkelson's (2002) technique. This approach was developed in the context of metacommunities, but can also be applied to species interactions (Lewinsohn et al. 2006). The analysis begins by reordering the observed binary matrix using reciprocal averaging (i.e. correspondence analysis). This ordination technique rearranges rows and columns to minimize ‘embedded absences’, namely interruptions in contiguous sequences of species interactions (cf. network ‘intervality’, see Stouffer et al. 2006). For example, a row comprised of three consecutive interactions (i.e. 1,1,1) has no embedded absences, while a row comprised of two interactions separated by the absence of an interaction (i.e. 1,0,1) has one embedded absence. To test the significance of the number of embedded absences in a matrix, the ‘1's and ‘0's in the observed matrix are randomly rearranged, reordinated and the number of embedded absences in the resulting null matrix is tallied. If the observed number of embedded absences is less than that found in 95% of null matrices (N = 500), the observed matrix is said to be ‘coherent’.
If a matrix is coherent, additional null models are then used to test for boundary clumping and nestedness. Boundary clumping is the tendency of coherent species interactions to coincide with one another and results in ‘compartmented’ species interactions (Lewinsohn et al. 2006). The method derived by Hoagland & Collins (1997), which employs a chi-squared test based on Morisita's index (Morisita 1971), is used to test for non-random boundary clumping in both host trees (rows or ‘communities’) and epiphytes (columns or ‘ranges’).
Leibold & Mikkelson (2002) characterize nestedness with species ‘replacements’, namely pairs of species that substitute each other's interactions, which is analogous to checkerboard distributions in island biogeography (Diamond 1975). This metric is inversely related to nestedness and a perfectly nested matrix has no species replacements. Two null models are used to test whether the observed number of species replacements differs from randomized expectations. The ‘community’ null model randomizes the positions of contiguous distributions of species interactions in each row of the observed matrix and calculates the resulting number of species replacements, while the ‘range’ null model randomizes the range of species interactions in each column. Evidence for nestedness is observed when over 95% of null matrices (N = 500) have more replacements than the observed matrix. All of Leibold & Mikkelson's (2002) null models were conducted using codes obtained from G.M. Mikkelson (personal communication).
Next, I used an assortment of previously derived null models that test for nestedness by randomizing the locations of species interactions (links) in the observed matrix. The nestedness temperature calculator (NTC), and its associated null model, is perhaps the most commonly used technique (Atmar & Patterson 1993). It characterizes nestedness with a metric called matrix temperature (T), which is a measure of disorder that is inversely related to nestedness (N). Matrix temperature is calculated by first reorganizing the matrix in order of increasing row and column sums. When the matrix is ‘maximally packed’ after reorganization, a curve (isocline) is fit through the matrix in a way that best separates occupied cells from unoccupied cells. Deviations from this curve are used to calculate T, which I then converted to N following Bascompte et al. (2003) (N = (100 − T) · 100 – 1). To evaluate whether the observed binary matrix differs from randomized expectations, the positions of entries in the matrix are randomly shuffled in a computer simulation, assuming an equal probability of drawing an interaction in each cell of the matrix. Differences between observed and expected levels of nestedness are used to establish whether the observed matrix is structured non-randomly. Five hundred simulations of the observed matrix were used to test for nestedness in epiphyte–host interactions using the NTC.
Rodríguez-Gironés & Santamaría (2006) identified several problems with the NTC: (i) the algorithm used to maximally pack the observed matrix does not always provide the best solution; (ii) the isocline reflecting a perfectly ordered matrix is not uniquely defined; and (iii) the null model used to establish deviations from randomized expectations underestimates type-1 error rates (see also Cook & Quinn 1998; Fischer & Lindenmayer 2002; Bascompte et al. 2003). Therefore, their technique, which corrects for these problems, was conducted using codes obtained from the corresponding author (M.A. Rodríguez-Gironés, personal communication). Similar software is also available from Guimarães & Guimarães (2006) (http://www.guimaraes.bio.br/sof.html). Five hundred simulations were conducted on the recommended settings (population size = 30, number of individuals = 7, generations = 200).
Two null models provided by the Rodríguez-Gironés & Santamaría (2006) software were used to test for nestedness. The first was originally derived by Fischer & Lindenmayer (2002) and was developed in the context of island biogeography. It randomizes the matrix according to the number of islands occupied by each species, but not island suitability. In other words, it considers column occupancy but not row occupancy. The second null model was that derived by Bascompte et al. (2003), which was developed in the context of mutualist networks. It randomizes the matrix according to the generalization level of both players in the mutualism, and therefore considers both row and column occupancies.
Lastly, I tested for nestedness by deriving a new null model that randomly populated host trees with epiphytes. Five hundred null interaction matrices were generated using the same procedure used in the degree distribution null model. All individual epiphyte occurrences observed in the field were randomly assigned to host tree species according to their overall frequencies of occurrence in the community. The resulting matrix was transformed into a binary interaction matrix, manually imported into the NTC and its level of nestedness was calculated. The level of nestedness in the observed matrix, which was also obtained in the NTC, was then compared with the distribution of nestedness values of null interaction matrices. The fraction of null matrices generating nestedness values greater than or equal to that observed was multiplied by two to obtain a two-tailed, type-one error rate for evidence of nestedness (i.e. P-value). If the observed level of nestedness is statistically indistinguishable from null interaction matrices, it would suggest that the observed level of nestedness results from the frequency of species interactions. On the other hand, if the observed level of nestedness is higher than that generated by this null model, it would indicate that processes other than the frequency of species interactions are important.
A total of 261 epiphyte species occurrences were encountered on 274 host trees belonging to seven species. Nineteen vascular plant species were found growing epiphytically (see Appendix S1). Hyperbolic functions provided a good fit to all rarefaction analyses relating the number of links to the number of individual epiphyte species occurrences observed on host trees (Fig. 1, average R2 = 0.996, range 0.989–0.999). Estimates of the sampling effort required to uncover a single additional link based on these functions indicated that additional linkages would only be encountered with substantial additional sampling. Dacrydium cupressinum and B. tawa would require modest amounts of additional sampling to acquire an additional link, 13% and 35%, respectively. Approximately half of the original effort would be needed for K. excelsa (50%), M. ramiflorus (55%) and P. ferruginea (80%). Double the sampling effort would be required for D. spectabile (175%), E. dentatus (137%) and the community as a whole (174%). Therefore, sampling efficiency appeared adequate to characterize the epiphyte metacommunity.
Epiphyte and host tree species showed strong differences in the number of times they were encountered during sampling (see Fig. 2). Both groups were characterized by a small number of species that were frequently encountered and a large number of species that were rarely encountered. The number of links associated with each epiphyte species was positively correlated with the number of individual trees on which it was found (product-moment correlation, R = 0.895, P < 0.001). Similarly, the number of links associated with each host species was positively correlated with the total number of epiphyte species occurrences it supported (R = 0.803, P = 0.030).
Degree distributions of epiphyte and host tree species matched that predicted by the null model (Fig. 3). The observed number of links for each epiphyte species was positively associated with null model predictions (n = 19, R2 = 0.838, P < 0.001). The y-intercept of the relationship (0.041) did not differ statistically from zero (95% CI = −0.800–0.882) and the slope (0.914) did not differ from one (95% CI = 0.715–1.112). The observed number of links for each host tree species was also positively related to null model predictions (n = 7, R2 = 0.739, P = 0.008). The y-intercept of the relationship (–6.650) did not differ statistically from zero (95% CI = −16.595–3.295) and the slope (1.598) did not differ from one (95% CI = 0.630–2.565).
Leibold & Mikkelson's (2002) technique indicated that the observed matrix was significantly coherent (P < 0.002). However, null model tests for boundary clumping yield mixed results. The ‘community’ null model indicated that interaction boundaries were neither clumped nor over-dispersed (P = 0.809), while the ‘range’ null model indicated that interaction boundaries were significantly clumped (P = 0.001). Leibold & Mikkelson's (2002) technique revealed strong, consistent support for nestedness. Both the ‘community’ (P < 0.002) and ‘range’ (P < 0.002) null models indicated that the observed matrix had fewer species replacements than expected by chance.
All other matrix reshuffling null model tests for nestedeness yielded similar results. The nestedness temperature calculator indicated the observed value of nestedness (N = 93.22) was significantly higher than randomly reshuffled matrices (Fig. 4). The binary matrix temperature calculator gave similar results (N = 94.15), with both of its null models indicating significant departures from randomly reshuffled matrices (Fischer & Lindenmayer's (2002) null model, P = 0.002; Bascompte et al.'s (2003) null model, P < 0.001).
The newly derived null model, which randomly generated null matrices from the frequency of species interactions also showed support for nestedness (Fig. 4). Only 11 out of 500 simulation replicates were more nested than the observed matrix (P = 0.044).
Degree distributions of epiphytes and their host trees were consistent with a null model that randomly populated host trees with epiphytes according to their overall frequencies of occurrence in the community. Congruence between observed and expected patterns suggests that there are few constraints on the development of degree distributions and that the number of links formed by each epiphyte and host species is determined by the frequency of species interactions. On the other hand, the species composition of epiphyte–host interactions differed strongly from null model expectations. Nestedness of the observed matrix was higher than several types of differently constructed null models, suggesting that the species composition of epiphyte–host interactions is structured deterministically.
The ‘nestedness hypothesis’ (sensu Ollerton et al. 2007) states that nested mutualisms are generated by the core of interacting generalist species, which provide a stable set of resources for the inclusion of more specialized species, which by definition only interact with generalists (Thompson 2005). A similar process may be at work in epiphyte communities. Host trees lack epiphytes in their early stages of growth and accumulate epiphytes through time. The most generalized epiphyte, Pyrrosia eleagnifolia, is typically the first to colonize host trees. It has a prostrate growth habit and forms dense vegetative mats that extend horizontally across branches. Small amounts of soil get trapped amongst their root systems and other species, such as ferns with more erect growth habits (e.g. Asplenium flaccidum, A. polyodon) and nest forming lilies (Collospermum hastatum, Astelia solandri), geminate in the interspaces of their root systems. These species further accelerate the process of soil accumulation, by failing to abscise old, dead leaves and trapping leaf litter produced by their hosts. This process ultimately produces large soil ‘nests’, which house diverse soil microbial communities (Wardle et al. 2003). Specialized epiphyte species, such as Dendrobium cunninghamii and Pittosporum cornifolium, in addition to species that sometimes grow epiphytically but normally occur on the forest floor, are nearly always found growing in these ‘nests’. Barren tree-tops are demanding places for plants to grow because of low moisture and nutrient availability, and specialist epiphyte species would probably be unable to exist within the metacommunity without the environmental changes brought about by earlier colonists. Consequently, nestedness in epiphyte–host interactions may result from the sequential colonization of epiphyte species that are less tolerant of soil moisture and nutrient stress (see also Stone 1989; Ellis & Coppins 2006). However, this explanation is speculative and awaits quantitative testing.
Jordano et al. (2003) analysed the degree distributions of a variety of pollination and seed dispersal mutualisms. They found that most mutualistic networks differed from the ‘scale-free’ properties predicted by models that preferentially attach links to highly connected species (Barabási & Albert 1999). They speculated that ‘forbidden links’ generated by phenotypic incompatibility between pairs of mutualist species were responsible for departures from scale-free topologies. For example, flower species with long corollas cannot be linked to insect pollinators with short proboscises (Stang et al. 2006), and winter-ripening fruit species growing at high latitudes cannot be dispersed by frugivore species that migrate to the tropics in winter (Burns 2006). However, Jordano et al. (2003) did not test whether observed degree distributions could be explained by the frequency of species interactions. They only tested for departures from scale-free topologies.
In subsequent null model analyses, Vázquez (2005) compared the same data with degree distributions that were randomly generated by a null model based on the frequency of species interactions. He found that observed degree distributions were consistent with null model expectations, suggesting that there are few constraints on the number of links maintained by each species. Similarly, Vázquez et al. (2005) documented that asymmetry in the degree of flea-mammal and metazoan-fish networks was consistent with their relative abundances, and Brooks et al. (2006) found that the number of platyhelminth parasites associated with tropical anurans is associated with the amount of time each anuran species spends in habitats where they are exposed to infection. Degree distributions of epiphytes and their host trees in New Zealand are consistent with these results. The number of links associated with both players in the interaction appears to be determined stochastically from differentially abundant species.
Null model tests for nestedness yielded different results. The species composition of epiphyte–host interactions was more nested than expected under several null models that randomly reshuffle observed species interactions. This result is consistent with most types of mutualistic interactions, including plant-pollinator (Phillip et al. 2006), fruit-frugivore (Lázaro et al. 2005), ant-plant (Guimarães et al. 2006), clownfish-anemone (Ollerton et al. 2007) and marine fish cleaning symbioses (Guimarães et al. 2007). In fact, the level of nestedness between epiphytes and their host trees ranks among the highest yet recorded for any type of species interaction.
Leibold & Mikkelson's (2002) technique yielded mixed support for compartmented interactions (sensu Lewinsohn et al. 2006). Although no evidence for boundary clumping was found for host trees, epiphyte boundaries were significantly clumped. Clumped epiphyte boundaries appear to result from the tendency of rare epiphyte species to occur only on E. dentatus, which had the largest number of links of any host species. This effect (i.e. boundary clumping in a nested interaction matrix with long ‘tails’) is graphically illustrated in Lewinsohn et al. (2006, Figure 1C and 1K). Therefore, while some evidence for compartmentalization was observed, it was consistent with a strongly nested topology.
Nestedness could result from unconstrained interactions between species that differ in relative abundances. Higgins et al. (2006) conducted computer simulations that randomly populated virtual island archipelagos with differentially abundant species. Most simulations generated nested patterns in island species composition and a similar process could be responsible for nested interaction networks. Assuming there are no constraints on species interactions, nestedness might result from differences in the frequency of interactions between each host and epiphyte species. However, results were inconsistent with the hypothesis that nestedness results solely from differences in interaction frequencies. The observed matrix was more nested than over 95% of all null matrices generated by randomly placing epiphyte occurrences onto host trees. Therefore, while the frequency of species interactions may contribute to the observed level of nestedness, it cannot explain it entirely, and deterministic processes appear to influence the species composition of epiphyte–host interactions.
Several recent studies have found that network properties can be explained by both the frequency of species interactions and forbidden links. Blüthgen et al. (2006) documented that pollination networks aren't always predicted by the relative abundances of mutualists, while Santamaría & Rodríguez-Gironés (2007) found that ‘mixed-models’, which incorporated the effects of both relative abundances and forbidden links, provide the best fit to many previously published pollinator networks. Similarly, Stang et al. (2007) found that both species’ abundances and forbidden links resulting from morphological mismatching explained the network structure of a Mediterranean pollinator mutualism.
Perhaps because of the logistical difficulties associated with studying epiphytes, ecological research on epiphytes often focuses on population or species-level processes. However, recent research has begun to explore broad-scale processes in epiphytes more intensely. For example, null models have recently been used to test for non-random patterns in host specificity (Laube & Zotz 2006) and to evaluate whether changes in epiphyte diversity along elevation gradients show mid-domain effects (Cardelús et al. 2006). Results reported here suggest that deterministic processes generate nested epiphyte–host species interactions and that nestedness of epiphyte–host interactions ranks among the highest yet recorded for any type of ecological interaction. Results also indicate that epiphyte assemblages are excellent examples of metacommunities (see Leibold et al. 2004), and when viewed from a metacommunity perspective, epiphytes could be ideal systems to test competing paradigms that attempt to account for ecological community structure.
I am grateful to John Dawson for his experience with the New Zealand flora and his willingness to share it with me. Diego Vázquez and two anonymous reviewers provided helpful comments on the manuscript. Thanks also to the staff at Otari-Wilton's bush for providing logistical support and Victoria University of Wellington for funding.