vol. 168, no. 1 the american naturalist july 2006 E-Article Phylogenetic Analysis of Trophic Associations A. R. Ives1,* and H. C. J. Godfray2,��� 1. Department of Zoology, University of Wisconsin, Madison, Wisconsin 53706 2. Natural Environment Research Council Centre for Population Biology, Division of Biology, Imperial College London, Silwood Park Campus, Ascot, Berkshire SL5 7PY, United Kingdom Submitted July 29, 2005 Accepted February 27, 2006 Electronically published May 19, 2006 Online enhancement: Matlab zip file. abstract: Ecologists frequently collect data on the patterns of as- sociation between adjacent trophic levels in the form of binary or quantitative food webs. Here, we develop statistical methods to es- timate the roles of consumer and resource phylogenies in explaining patterns of consumer-resource association. We use these methods to ask whether closely related consumer species are more likely to attack the same resource species and whether closely related resource species are more likely to be attacked by the same consumer species. We then show how to use estimates of phylogenetic signals to predict novel consumer-resource associations solely from the phylogenetic position of species for which no other (or only partial) data are available. Finally, we show how to combine phylogenetic information with information about species��� ecological characteristics and life- history traits to estimate the effects of species traits on consumer- resource associations while accounting for phylogenies. We illustrate these techniques using a food web comprising species of parasitoids, leaf-mining moths, and their host plants. Keywords: host-parasitoid interactions, community structure, predator-prey associations, phylogeny, comparative methods. Why are some plant species attacked by many more species of herbivore than others (Strong et al. 1984)? And why do some parasite species have a much broader host range than more specialized species (Price 1980 Combes 2001)? These are examples of questions that can be asked of data sets that describe how two adjacent trophic levels in a food web are linked together. Not all food webs have species * Corresponding author e-mail: arives@wisc.edu. ��� E-mail: c.godfray@imperial.ac.uk. Am. Nat. 2006. Vol. 168, pp. E1���E14. 2006 by The University of Chicago. 0003-0147/2006/16801-41209$15.00. All rights reserved. organized in distinct trophic levels, but many do, or do to a good approximation, and increasingly quantitative data are becoming available that describe large sets of in- teracting species (van Veen et al. 2005). Most analyses of these types of questions have either assumed that data from different species are statistically independent or considered the possibility of phylogenetic correlation at one trophic level but not the other. As is now widely appreciated, there is a danger of assuming that data from species are statistically independent closely re- lated species may share traits simply because they have inherited them from a common ancestor (Felsenstein 1985 Harvey and Pagel 1991). For example, swallows and martins (Hirundinae) reuse their nests year after year and are attacked by a relatively large number of species of insect ectoparasites (Rothschild and Clay 1952). However, this may have nothing to do with nesting habits and just be due to the common lousy ancestor of the clade some other unmeasured trait might cause the high ectoparasite bur- den, with all swallows and martins sharing this unfortunate trait due to their common phylogenetic history. As Charles Darwin pointed out, the traits of species often reflect their evolutionary history, and when species traits affect the range of hosts they attack or the range of parasites that attack them, phylogenetic patterns of host-parasite asso- ciations are inevitable. For analyses restricted to one trophic level, a wide range of statistical techniques is now available to estimate the extent of phylogenetic signal and to accommodate phy- logenetic correlation in hypothesis testing. There are, how- ever, few methods that simultaneously allow phylogenetic information from both trophic levels to be incorporated into the analysis of patterns of association. Although Page and colleagues (Page 1994, 2000 Charleston 1998 Page et al. 1998) developed methods to detect cocladogenesis (cospeciation) in ectoparasites of birds and similar prob- lems (Charleston 1998), here the parasites are monoph- agous or restricted to small clades of hosts, and phylo- genetic processes dominate. Their techniques are not designed to explore patterns in more connected assem- blages, where ecological determinants of trophic links may be as strong as or stronger than the effects of phylogeny. The aim of this article is to develop methods for as-

E2 The American Naturalist sessing the strength of phylogenetic signal that determines the patterns of association between consumers and re- sources in simple food webs. Specifically, we ask whether a given resource species is more likely to be used by phy- logenetically related species of consumers and whether a given consumer species is more likely to use phylogenet- ically related resource species. Our objective is to identify the presence of phylogenetic signal (Blomberg et al. 2003) without attempting to ascribe any adaptive explanation identifying phylogenetic correlations is a statistical issue, whereas ascribing adaptive explanations goes far beyond statistics (Blomberg and Garland 2002). We also ask whether it is possible to predict how a novel introduced species will fit into the food web, based on its phylogenetic position. Our motivation here is to help predict which native resource species will probably be at risk from an invasive consumer and which native consumers may be effective control agents for invasive resource species. Fi- nally, we develop methods for regression analyses in which the strength of consumer-resource associations is regressed on life-history or environmental factors of consumer and/ or resource species while incorporating phylogenetic sig- nals. Incorporating phylogenetic signals into regression analyses will lead to the most precise estimates of regres- sion coefficients. Throughout, we take a statistical ap- proach, so parameter estimates and predictions are made with confidence intervals. To illustrate and test the techniques for analyzing the cophylogenies of resources and consumers, we explore the patterns of association in a quantitative host-parasitoid food web. Parasitoids are insects that lay their eggs in or on the bodies of other insects, their hosts, which provide all the resources the developing parasitoid needs to reach maturity (Godfray 1994 Quicke 1997). Because of para- sitoids��� importance in biological control, there are abun- dant data on the parasitoid species complement of differ- ent hosts, and fewer but still numerous data on the number of host species attacked by different parasitoids. Analysis, largely without incorporating phylogenetic information, has suggested a variety of possible patterns, such as the effect of host geographical range on the number of par- asitoid species that attack them and the effect of parasitoid feeding strategy (internally or externally on the host) on the host species they attack (Hawkins 1994). Recently, a number of quantitative host-parasitoid food webs have been constructed in which not only are the suitable hosts established for each parasitoid species but the strength of interaction between each host-parasitoid pair is quantified by the parasitism rates (e.g., Mu ��ller et al. 1999 Lewis et al. 2002). These quantitative food webs can be used to explore community structure and the potential for indirect interactions. Because these webs describe the densities of hosts, parasitoids, and their associations in the same units, they are ideal for the quantitative exploration of the strength of phylogenetic signal in determining food web structure. The food web we study includes 12 leaf-mining moths and 27 species of parasitoid wasps. We first ask whether there is phylogenetic signal in the observed associations between hosts and parasitoids and whether this signal acts through the host and/or parasitoid phylogeny. In addition to considering the phylogeny of the leaf miners, we also ask whether the host-parasitoid associations could be ex- plained by the phylogeny of the plant species used by the leaf miners. This would occur if the parasitoid host range and foraging strategy are determined, at least in part, by the plant species used by the leaf miners (Vinson 1997). Once parameters for the phylogenetic signals are esti- mated, it is possible to use the phylogenies (discounted by the estimates of signal strength) to predict the host range for a novel parasitoid from the parasitoid���s location on its phylogenetic tree and to predict the potential par- asitoids of a novel host from the host���s location on its phylogenetic tree. Finally, we consider factors that may influence the susceptibility of hosts to parasitism and fac- tors that may influence the efficacy of parasitoids searching for hosts. Specifically, we ask whether hosts with broader geographical ranges experience parasitism from a greater number of parasitoid species (Hawkins and Lawton 1987) and whether parasitoids that feed externally on hosts (ec- toparasitoids) attack a greater number of host species than those that feed internally on still-living hosts (endopar- asitoids Askew and Shaw 1986 Sheehan and Hawkins 1991). Methods Comparative Methods for Interacting Trophic Levels Consider information on the strengths of association be- tween n resource species and m consumer species in a food web or similar data structure. For narrative simplicity, we henceforth call the resources ���hosts��� and the consumers ���parasitoids.��� Let the association strengths (whose exact form we need not specify for the moment) be denoted Aik ( The total number of asso- i p 1, ��� , n k p 1, ��� , m). ciations (including zeros) is N p nm. Assume that there is a host trait Xh and a parasitoid trait X p that together affect the hosts��� susceptibility to par- asitism and the parasitoids��� effectiveness at finding and attacking hosts. These traits are assumed to depend on phylogenetic relatedness so that closely related host or par- asitoid species are more likely to have similar values of Xh or X p , respectively. We assume that phylogenetic related- ness gives rise to covariance matrices V and U, whose elements and ukl describe the covariance in the trait vij