Multitrophic interactions play key roles in the origin and maintenance of species diversity, and the study of these interactions has contributed to important theoretical advances in ecology and evolutionary biology. Nevertheless, most biodiversity inventories focus on static species lists, and prominent theories of diversity still ignore trophic interactions. The lack of a simple interaction metric that is analogous to species richness is one reason why diversity of interactions is not examined as a response or predictor variable in diversity studies. Using plant–herbivore–enemy trophic chains as an example, we develop a simple metric of diversity in which richness, diversity indices (e.g., Simpson's 1/D), and rarefaction diversity are calculated with links as the basic unit rather than species. Interactions include all two-link (herbivore–plant and enemy–herbivore) and three-link (enemy–herbivore–plant) chains found in a study unit. This metric is different from other indices, such as traditional diversity measures, connectivity and interaction diversity in food-web studies, and the diversity of interaction index in behavioral studies, and it is easier to compute. Using this approach to studying diversity provides novel insight into debates about neutrality and correlations between diversity, stability, productivity, and ecosystem services.
Las interacciones multitróficas son en gran parte responsables por el origen y el mantenimiento de la diversidad de especies, y las investigaciones de estas interacciones han contribuido a avances teóricos en ecología y biología evolutiva. Sin embargo, en la mayoría de los inventarios de biodiversidad se hacen listas estáticas de especies, y las teorías de la diversidad más prominentes siguen ignorando las interacciones tróficas. La falta de una métrica de interacción sencilla y que sea análoga a la riqueza de especies es una de las razones para que no se examine la diversidad de interacciones como una respuesta o una variable dependiente en estudios de diversidad. Usamos las cadenas tróficas de plantas-herbívoros-enemigos como un ejemplo y desarrollamos una variable simple de diversidad en la cual se puede calcular la riqueza de interacciones o los índices de diversidad (por ejemplo, 1 / D de Simpson) y la diversidad de rarefacción; se usan los enlaces como la unidad básica en lugar de las especies. Las interacciones incluyen todas las interacciones de dos enlaces (planta-herbívoro y herbívoro-enemigo) y de tres enlaces (enemigo-herbívoro-planta) que se encuentra en muestreos o censos por transectos o por parcelas. Esta variable es diferente de otros índices de diversidad tradicionalmente usados como la conectividad de interacción en estudios de cadenas tróficas, y el índice de diversidad de interacciones que se usa en estudios comportamentales, además es más fácil de calcular. Usando esta aproximación para estudiar la diversidad de interacciones creará nuevas perspectivas en debates sobre la teoría neutral y las correlaciones entre la diversidad, la estabilidad, la productividad, y servicios de los ecosistemas.
Understanding the origin and maintenance of tropical diversity has always been one of the major goals of tropical ecology. Considerable progress has been made toward integrating the many feasible and well-tested hypotheses for why there is such a strong latitudinal gradient in diversity for most taxa (Willig et al. 2003, Mittelbach et al. 2007, Arita & Vazquez-Dominguez 2008). Nevertheless, there are important unanswered questions about how patterns of diversity should be measured, catalogued, monitored, and conserved (Reiss et al. 2009). In diverse tropical habitats, efforts to document species richness are naturally stymied by overwhelming biodiversity and rapid anthropogenic degradation of habitat. Tropical areas suffer disproportionate losses in biodiversity, at a rate measured in weeks rather than years, and there is an urgent need to develop and implement novel approaches to understanding tropical ecosystem structure.
A rich literature exists for interaction diversity, which was loosely defined by Janzen (1974) and then by Thompson (1996) as the number of interactions linking species together into dynamic biotic communities. Janzen lamented the loss of interactions as a hidden consequence of extinction and Thompson (1996, 1997) pointed out that most of the earth's diversity is in the interaction diversity found in the tropics (also see Ohgushi et al. 2007). Numerous studies have acknowledged that major theoretical and applied issues in ecology and conservation must go beyond documenting species diversity and take into account interaction diversity (Cohen & Briand 1984, Price 2002, Del-Claro 2004, Burslem et al. 2005, Ohgushi et al. 2007, Tylianakis et al. 2007, Del-Claro & Torezan-Silingardi 2009). All these studies acknowledge that interactions among species are important because they affect all ecosystem attributes, from primary productivity to population dynamics and the survival and reproduction of individuals. Nevertheless, there has been little effort to quantify interaction diversity in the same way that species diversity has been catalogued, mapped, highlighted, and used as a predictor or response variable in ecological studies (Ollerton & Cranmer 2002).
There are a number of methods available for measuring interaction diversity. The most prominent are found in literature on food webs and in behavioral studies (Slater 2000, Bersier et al. 2002). Quantitative food webs are an improvement over qualitative (binary) food webs because predators and prey are measured in the same units of density, allowing the relative differences in the number of linkages observed for each species to be proportional to their actual frequency in the community. In a quantitative food web, the abundance of each species is measured as a density such that the observed species abundances are assumed to be a representation of the populations of each species and their corresponding linkages in the community. In these studies, sampling variance, which affects the number of individuals of each species and thus the number of observed linkages, is the major source of unexplained variance and may be accounted for using a variety of subsampling and data exclusion methods.
Quantitative food-web studies typically assess food-web structure using measures of connectance and compartmentalization. Connectance statistics assess the degree of species-level specialization between trophic levels by comparing the number of observed linkages to the number of potential linkages in a completely connected (generalist) food web. Low measures of connectance indicate communities with highly specialized consumers (Bersier et al. 2002). Graphic representations of food webs include visual representations of the relative abundance of species and linkages, but the information in these figures is not easily summarized statistically. Compartmentalization statistics measure the equivalent number of species within each independent group of interacting species. Here pi is measured as the fraction of total species observed in a single compartment and species abundance is ignored. This measure is useful for determining the degree to which interacting species are divided into independent ecological units, and is valuable for designing experimental approaches to communities or community subsets. The number and evenness of compartments, however, assess a narrow aspect of trophic interactions that is maximized only when compartments are equivalent in their number of species. There remains no generalized analytical approach to measuring the structure of species interactions that includes information regarding the richness and abundance of individuals and linkages in a community.
A new focus in food-web studies and interaction diversity is to examine nestedness (Bascompte et al. 2003) and this is an effective method for examining interaction diversity. Qualitative mutualistic webs (e.g., pollination, seed dispersal, and protection; Bascompte et al. 2003, Burns 2007, Ollerton et al. 2007, Rezende et al. 2007) have been examined for the degree of nestedness. One interesting pattern emerging from this work is that these mutualistic networks are cohesive in that the generalists in these systems create a dense core of interactions that provides a foundation for the rest of the community (Bascompte et al. 2003). This approach allows for insight about the consequences of losses of specific interactions or species from a community.
Unfortunately, these food web and network approaches involve extensive knowledge of natural history, which is not completely available for any ecosystem (Godfray et al. 1999, Novotny & Basset 2005), especially in the tropics. Although some well-resolved food webs exist (Kelly & Southwood 1999, Memmott et al. 2000, Brandle & Brandl 2001), these are merely slices embedded in more complex webs of interaction. This is true for even ubiquitous, cosmopolitan, and well-studied systems such as alfalfa monocultures, where species interactions are still not fully resolved (Dyer & Stireman 2003, Pearson et al. 2008). This lack of actual food-web data illuminates a major problem facing current diversity research: accumulation of relevant empirical data lags far behind the plethora of mathematical, graphical, and verbal models in ecology (Dyer 2008). Thus, we cannot test hypothesized correlations between tropical diversity and variables such as specialization, predation, and competition because the growing literature on ecological theory relies on a large number of untested assumptions, such as the concept that tropical organisms are more specialized.
In the tropics, our knowledge of the basic life histories for plants and animals (e.g., Singer et al. 1982, Fortey 2008, Stork 2008) remains in its infancy. Rapidly accumulating species lists provide a blurry glimpse of community assemblages, but their lack of information on community structure provides only weak suggestions for conservationists and those attempting to preserve them or counter the effects of habitat fragmentation and its disruption of ecological processes. The current devaluation of natural history studies has been lamented by many modern biologists and their importance touted by many others (Futuyma 1998, Dayton 2003), but a demand for novel theoretical advances seems to have eclipsed any coordinated efforts for basic descriptions of ecological communities. We suggest that using subsamples of life-history strategies (i.e., interactions) will allow for quantification of natural history observations and can provide unique insight into the understanding of patterns of biodiversity.
For tropical ecosystems, the lack of a simple interaction diversity metric or approach is exacerbated by the fact that most empirical and many theoretical studies of diversity are potentially irrelevant to tropical ecosystems given the limited range of diversity studied. For example, the classic studies of Tilman (e.g., Tilman et al. 2006) examine effects of 1–24 species of plants on primary productivity or ecosystem stability, whereas changes in plant richness from temperate to tropical ecosystems or from low to high elevation can be two to three orders of magnitude (Kreft & Jetz 2007). Similarly, food-web studies that range from 3 to 33 species (Cohen & Briand 1984) or mathematical models that compare two-predator systems to three-predator systems (e.g., Begon et al. 1996) are of limited utility when trying to understand causes and consequences of enemy diversity gradients that cross orders of magnitude. Thus, it is not clear how variables such as community stability and ecosystem function are affected when a forest patch has 1000 vs. 10 species of parasitoids or when a forest has 20 vs. 200 species of vascular plants.
A SAMPLING APPROACH TO MEASURING INTERACTION DIVERSITY
Ideally, a quantitative approach to measuring the properties of food webs should include quantified information on trophic interactions for all trophic levels and involves all species in a community, but such an approach is impractical for most reasonably diverse communities (Lewis et al. 2002). Quantifying diversity of interactions is similar to the measurement of species diversity and the numerous problems associated with measuring tropical diversity (discussed or reviewed by Heck et al. 1975; Lande 1996; DeVries et al. 1999, 2001; Magurran 2004; Jost 2006, 2007) apply to measures of interaction diversity (Cohen & Briand 1984). Because interactions are defined by the species observed interacting, distributions are likely to behave similarly to the number of species in having an effectively infinite number of unique units, even though many of those units are likely to be very rare or infrequent in samples. In addition, the number of interactions observed in a given sample is likely to be correlated with sampling effort.
The quantitative investigation of food webs therefore requires the development of sampling schemes designed to infer the properties of the complete community food web from samples that in practice represent only a portion of the community. Current quantitative food-web investigations solve this problem by intensively sampling a restricted food web, focused on guilds or subsets of interacting species (Memmott & Godfray 1994). This approach gains precise quantitative characterization of trophic interactions in terms of the numerical importance of interactions within the guild, but sacrifices inference to the other species in the community that were not sampled. An alternative approach is to sample interactions at random from the total community pool of interactions with the intent of estimating the properties of the total community food web. A clear advantage of this approach is that it allows inferences about the total community food web to be made directly from the information in samples because the experimental unit of interest is the total community as opposed to a single family or guild of organisms. By design this approach sacrifices the precision and within-guild resolution of interactions provided by exhaustive sampling of focal groups to gain greater insight for the entire community. The random sample approach to the study of food webs takes advantage of the well-established methods used in the study of species diversity to estimate the properties of the total community from a limited number of samples.
AN EXAMPLE: TRITROPHIC INTERACTIONS, DIVERSITY, AND SPECIALIZATION
Plants, phytophagous insects, and enemies make up well over half of global terrestrial biodiversity (Price 2002) and interactions between these trophic levels have generated important theoretical advances in ecology and evolutionary biology (Hairston et al. 1960, Ehrlich & Raven 1964, Lawton & McNeill 1979, Price et al. 1980, Novotny et al. 2006), but complete quantitative food webs including these guilds are scarce (Godfray et al. 1999). In fact, detailed information will never be available for all of the terrestrial food webs on earth, nor will any complete food web be constructed for a tropical ecosystem. Tritrophic chains are exemplary systems for creating food web metrics that broadly capture interacting species in a community but do not require documenting all the existing interactions. Using plant–herbivore–enemy trophic chains as an example, we recommend that tropical community ecologists utilize a simple metric of diversity in which richness, diversity indices, and rarefaction diversity are calculated with links as the basic unit rather than species. Interactions can include all two-link (herbivore–plant and enemy–herbivore), and three-link (enemy–herbivore–plant) chains found in a reasonable study unit, such as a 100-m2 plot or a 200-m transect (Fig. 1). The simplest measure is the total number of these interactions per unit area.
This sampling method depends on certain assumptions when analyzing interaction diversity as a response variable in models that compare effects between different habitats, elevations, and latitudes. If one samples a plot or transect for caterpillars, rears them out to yield an adult moth or parasitoid, and then counts up the interactions (Fig. 1), the distribution of unique interaction links can be analyzed using methods applied to traditional species diversity. For such an approach, we make four basic assumptions: (1) interactions are sampled at random, (2) interactions are homogenously distributed within a plot, (3) any distribution of abundance among interactions is possible, and (4) interactions are equally likely to be encountered at all locations compared.
We are currently utilizing this method to document interaction diversity at a number of tropical sites, including Yanayacu Biological Station Ecuador and La Selva Biological Station, Costa Rica, as well as temperate sites in the Sierra Nevada (California), Great Basin (Nevada), Arizona, and Louisiana. Plants and associated lepidopteran larvae and their parasitoids are quantified at all sites in 10-m diam. plots (methods described in Dyer et al. 2007). Preliminary analyses of plot data from the Ecuador and Louisiana sites, for which we have sufficient number of plots for initial analyses, indicate that interaction richness and diversity are significantly higher at the Ecuador vs. Louisiana site (for interaction richness, t=8.8, df=650, P<0.0001; Table 1), while species richness per unit area is higher at the Louisiana site. This is consistent with the hypothesis that interaction richness asymptotes more quickly than species richness (Fig. 1G), much like diversity indices asymptote more quickly than richness (Lande 1996). Although unique interactions are likely to be characterized by a lognormal distribution, this distribution should be hyperkurtotic, with a few common interactions dominating and most other interactions being extremely rare (Bascompte et al. 2003).
Table 1. Preliminary data comparing tropical (Ecuador) and temperate (Louisiana) interaction diversity in 10-m diam. circular plots. Means are reported (±SE).
Sample size (number of plots)
Mean species richness per plot
Mean interaction richness per plot
Mean interaction diversity (1/D) per plot
Measures of specialization are wrought with pitfalls (Novotny & Basset 2000, Ødegaard 2000, Novotny et al. 2006), but interaction diversity per unit area may be less susceptible to methodological flaws for measuring diet breadth as long as samples are random and well replicated (Fig. 2A). Thus, it is interesting to contemplate how interaction diversity per unit area, as well as beta diversity of interactions, would vary across a gradient from Europe to Papua New Guinea (Novotny et al. 2006) or across the Americas (Dyer et al. 2007; Figs. 2B, C, and E). The beta diversity, or spatial turnover, of species in tropical forests has recently become a focal point of analysis in the field of tropical ecology (Condit et al. 2002, Ødegaard 2006, Novotny et al. 2007, Lewinsohn & Roslin 2008). This concept of beta diversity or turnover could be expanded to examine how the diversity of interactions varies over space. Examination of interaction beta diversity could be used to determine whether species interact with the same set of consumers or hosts over their range and how rapidly these interactions undergo spatial turnover. Recent progress has been made by Novotny (2009) in developing a conceptual and analytical framework to quantify how diversity of interactions varies in space using a beta-diversity partitioning approach. In addition to the usefulness of such analyses in understanding patterns of community structure in a spatial context, they could aid in understanding important evolutionary phenomena, such as coevolutionary processes occurring over a geographic mosaic of interactions (Gomulkiewicz et al. 2000, Thompson 2005).
IMPORTANT QUESTIONS FOR INTERACTION DIVERSITY IN THE TROPICS
How is interaction diversity related to the important questions in biodiversity research? For many relationships, such as correlations between biodiversity, stability, and ecosystem function, interaction diversity is a more intuitive variable for understanding mechanistic relationships (Reiss et al. 2009). Thus, when Terborgh et al. (2001) showed that losing predator–prey interactions led to ‘ecological meltdown’ in islands, changes in the total number of species did not explain what happened to the island communities. It was clear that the response was due to losing important predator–prey–plant interactions—or a decline in interaction diversity. Stability can be understood in the same way—it is easier to explain or model the consequences of loss of interaction diversity than it is to understand the loss of species diversity (Fig. 2D). Much of the historical and current debate concerning the relationship between diversity and stability of ecological communities is concerned with the number of trophic links (i.e., connectance) and their relative strength (MacArthur 1955, May 1973, McCann 2000). For example, recent generalized modeling analyses of vast numbers of simulated food webs by Gross et al. (2009) have concluded that stability is enhanced by oligophagy in top predators and high diversity of intermediate trophic levels, and it is decreased by high variation in link (interaction) strengths. In addition, it has been argued that diverse weak interactions (‘the weak interaction effect’; McCann et al. 1998, McCann 2000) and trophic complexity (Jiang et al. 2009) both play pivotal roles in community stability. These relationships can readily be explored by examining interaction diversities between component trophic levels, even when complete food-web information is unavailable. Such interaction diversities and associated metrics can be compared among ecological communities or across gradients to make predictions concerning patterns of stability and resilience.
Understanding mechanistic relationships due to interaction diversity is also important for applied issues, such as determining the effects of climate change on biodiversity. Stireman et al. (2005) found that parasitoid abundance declines with increasing climatic variability. The mechanism proposed for this connection is related to interaction diversity. Extreme weather events may decrease interaction diversity via disrupting associations between parasitoids and their herbivorous hosts, thus decreasing the number of potential tritrophic interactions. In this case, it was the more specialized hymenopteran parasitoids that were most affected by increased climatic variability, suggesting that parasitoid–host interaction richness, as well as abundance, was reduced. Climate change and climatic variability have also been shown to influence diversity and trophic interactions in other systems, such as vertebrate–insect interactions (Spiller & Schoener 2008, Mazia et al. 2009) and invertebrate marine communities (Przeslawski et al. 2008). Extending analyses of these effects beyond species diversity and abundances by examining interaction diversity (Fig. 2F) may provide considerable insight into both the community consequences of climate change as well as the mechanisms involved.
Perhaps the most exciting change in biodiversity research for tropical ecology is the explosion of empirical and theoretical advances spurred by the development of the neutral theory (Hubbell 2001, Chave & Norden 2007, Zhou & Zhang 2008, Ellwood et al. 2009). Although neutral theory is relevant and useful for studying tropical diversity in theoretical frameworks by providing appropriate null hypotheses, it is primarily focused on single trophic levels. The majority of studies seeking to test if community structure can be explained by purely neutral processes have examined abundance distributions and diversity of plant species without regard to their interactions with consumers (Hubbell 2001, Chave 2004). Indeed, one problem with neutral theory as an explanatory model (rather than just a null model) is that it assumes that interactions among trophic levels are inconsequential in shaping species abundance distributions, despite extensive empirical data to the contrary (e.g., Paine 1969, Terborgh et al. 2001). Nevertheless, a similar ‘neutral’ approach can be used to develop models of interaction diversity based on the richness and abundance of trophically interacting communities. Such models can be used to assess empirical deviations from expected neutral ‘interaction distributions’ to understand how community structures differ from neutrality and provide some insight into why they may differ. This general approach of using neutral models to examine species interactions has proven profitable in recent studies on the asymmetry (Vazquez et al. 2007) and nestedness (Krishna et al. 2008) of mutualistic interactions. There is much room, however, for the development of null models that explain the structure of other types of trophic interactions, other properties of interaction diversity, and higher-order (e.g., tritrophic) interactions.
Finally, the latitudinal gradient in diversity is a robust and fascinating pattern of diversity that may be better understood by examining interactions. Novotny et al. (2006) utilized an indirect measure of interaction diversity to address a difficult question related to the diversity gradient: why are there so many species of tropical insects? These authors compared the number of herbivore species collected in standardized samples of tree species in tropical and temperate forests. They found that herbivore richness per tree species did not differ significantly across sites and that dietary specialization of herbivores (relative to the set of host species examined) was also similar for the two sites. Furthermore, they found that beta diversity across the lowlands of Papua New Guinea was relatively low (Novotny et al. 2007). Although they concluded that the latitudinal gradient in herbivorous insect diversity can be explained as a simple function of greater host plant richness in the tropics, a similar study indirectly examining interaction diversity across eight New World sites (Dyer et al. 2007) demonstrated that Lepidoptera are more specialized in the tropics than at temperate latitudes. Dyer and colleagues argue that between-plant species beta diversity in tropical lepidopteran herbivore communities is greater than in temperate communities, and that this may play an important role in shaping the latitudinal diversity gradient of insects. In both of these alternative views, the basic precept that ‘diversity begets diversity’ (Janz et al. 2006) is implicit; they differ rather in the form of that relationship—whether it is linear or nonlinear and how strongly it is influenced by other factors. More explicit and thorough analyses of interaction diversity across latitudinal gradients could help to determine more precisely how community structure varies latitudinally, how this variation contributes to patterns of species diversity, and help to reconcile these seemingly contradictory patterns (Figs. 2A–E).
It is clear that multitrophic interactions play key roles in the origin and maintenance of species diversity (e.g., Singer & Stireman 2005), thus it is relevant to utilize a biodiversity metric that quantifies these interactions. Current experiments that manipulate diversity via changing number of species, number of guilds/functional groups, identity of species, or other traditional diversity metrics should be supplemented with experiments that randomly assign different levels of interaction diversity to plots and examine the responses of productivity, mortality, temporal variance of species diversity, and other relevant response variables. Conservation issues cannot wait for a complete understanding of tropical life histories and community assemblages. Our proposed interaction diversity metric should form a basis for developing more realistic pictures of community structure for comparative purposes applicable to a broad array of ecological, evolutionary, and conservation issues. Marrying natural history, ecology, and conservation through the quantified sampling of interaction diversity is one-way forward to understanding the evolution, maintenance, and preservation of rapidly dwindling biodiversity.
This study was funded in part by NSF grants DEB-0346729, DEB-0527441, DEB-0849361, RAND Gulf States Policy Institute, National Geographic, and EarthWatch Institute. We thank G. Gentry for introducing us to the idea of interaction diversity and J. Ghazoul for inviting us to write this manuscript. Thanks also to A. Smilanich, M. Almeida Neto, and two anonymous reviewers for comments on the manuscript and to many other ecologists for discussions on these issues. Thanks to G. Gentry, W. Simbaña, R. Granizo, M. Ryan, J. Hazen, K. Hazen, M. Fox, T. Massad, and dozens of EarthWatch volunteers for conducting and supervising data collection. This is publication #203 of the Yanayacu Natural History Research Group.