What does it take to trace an entire food web? A colossal amount of work, to be sure. Whether aquatic or terrestrial, temperate or tropical, extensive food web studies are scarce and for good reasons. Well-known food webs tend to concentrate on larger organisms, and they exclude most smaller and more basal species, or else lump those into so-called trophic species (Dunne 2006) which can be comprised of a group of related genera up to an entire phylum. A different option is to focus on one particular interactive interface and distinguish species as fully as possible. This has become the preferred approach for studies of mutualistic webs encompassing plants and their floral visitors or fruit dispersers (Bascompte & Jordano 2007). Such highly resolved partial webs have opened new fronts for recognizing patterns in ecological assemblages and illuminating their organizing processes (Bascompte, Jordano & Olesen 2006; Vázquez et al. 2009).
In antagonistic or trophic webs, the same approach is advancing, although at a slower pace (Prado & Lewinsohn 2004; Thébault & Fontaine 2010). A recent In Focus piece (Beckerman & Petchey 2009) highlighted the importance of animal parasites in structuring the trophic interactions within the arctic food web studied by Amundsen et al. (2009). Ecologically, herbivorous insects also count as parasites of their host plants (Price 1980). In current estimates, they represent the largest share of terrestrial multicellular species (Foottit & Adler 2009). Therefore, to improve our understanding of the processes that organize and shape terrestrial biodiversity, the plant–herbivore interface is a prime centre of attention. Host specialization is potentially a major component of total herbivore diversity (Lewinsohn & Roslin 2008).
As a rule, plant–herbivore community studies focus on a particular feeding mode or animal taxon, which can be assessed through a single sampling or censusing method (e.g. Janzen et al. 2005; Ødegaard 2006). Novotny et al. (2010) advance further: they tackle the entire array of herbivorous insects associated with plants in a lowland rainforest area in Papua New Guinea. To accomplish this, they had to employ a battery of different procedures in the field, ranging from hand-picking of insects on leaves, to logging trees to provide oviposition substrates for wood-boring species. Herbivores were grouped into 11 guilds in a classification based on the combination of feeding mode (chewers or suckers), life stage (larvae or adults) and as external or internal feeders on various plant parts. In insects with incomplete metamorphosis, e.g. bugs and grasshoppers, nymphs and adults could be merged into the same guilds, whereas in fully metamorphosing groups (butterflies, beetles and others), larvae and adults often form distinct guilds.
In all, 224 plant species were studied, but different species were assessed for each herbivorous guild. To compare specificity and host patterns among guilds, the herbivores in each guild were analysed on sets of nine plant species that conformed to a standard taxonomic arrangement. Guilds separate into two markedly distinct groups. Leaf-mining, leaf-chewing and fruit-chewing larvae, plus leaf-sucking Heteroptera, are dominated by highly specialized species; most of whom are monophages within the standardized nine-host set. Sap-suckers (both phloem and xylem), root and stem-borers (including fungus eaters) and adult leaf-chewers are more generalistic, their members feeding on 1.7–3 hosts within the standard set of nine plant species.
These differences among guilds show up strikingly in the corresponding trophic webs. For the specialist guilds, webs are mostly split into semi-isolated compartments (Prado & Lewinsohn 2004), which correspond to source webs (sensuCohen 1978). On the other hand, in generalist guilds, herbivores broadly overlap among hosts. The more specialized guilds are likely candidates for density-dependent interactions with their hosts (Strauss & Zangerl 2002), with few effects on other plants or their herbivores. Generalist herbivores can mediate apparent competition among spatially and phylogenetically separate hosts. Generalistic guilds are therefore more liable to propagate effects among different host plants and also among their respective source webs (Lewinsohn, Novotny & Basset 2005).
Herbivorous guilds differ in their impacts on hosts not just because of their specialization. Novotny et al. note that wood-borers that feed on dead or dying tissue are unlikely to impact their host populations. They also include fruit-chewing insects in the same case, but this is because of the restricted coverage of that particular guild in their studies (only fruit-fly larvae feeding in fleshy pericarps were sampled). This under-represents fruit-feeding guilds, both in diversity and in importance. In fact, in accordance with the criteria with which they define guilds, pericarp-chewing or sucking herbivores should be separated from proper seed-feeders (Janzen 1980). The former could affect seed fate by interfering with their development, dispersal or germination; the latter form a distinctive set of organisms which can, and often do, affect plant demography directly (Strauss & Zangerl 2002). Similarly, one might go on to separate meristem-chewers from other herbivores given the disproportionate effect that meristem destruction can have on the survival and growth form of host plants. But then, if the devil lies in the details, ‘improving’ any classification must be one of his favourite pastimes.
Beyond the comparison among guilds, Novotny et al. attempted to estimate the size of the entire local plant–herbivore assemblage. Another complicated task, this entailed extrapolating species and interactions not recorded in the sampled guilds, plus completing gaps for other guilds based on studies conducted elsewhere. The entire dimension of the plant–herbivore web comes to nearly 10 000 insect species on about 200 host plants, with an estimated grand total of 50 000 trophic interactions; only 15% of the insects and links were actually recorded in the study.
There is little opportunity to contrast these findings to other studies; not only are other studies scarce, they are not generally comparable. Reagan & Waide (1996) summarized the entire terrestrial food web of a Puerto Rican forest – a long-term study involving many researchers. This encompasses vertebrates and invertebrates, herbivores, predators and scavengers. To summarize the web, their ‘maximum resolution version’ recognizes 2600 trophospecies, with about 59 000 links. At first sight, numbers are of the same order of magnitude as in Novotny and colleagues’ results; moreover, in that Puerto Rican forest, the plant diversity is similar to the forest sample in Papua New Guinea – 214 species. However, for lack of detailed herbivore feeding records, Reagan & Waide (1996) chose to class all plant resources into purely structural categories (e.g. leaves, seeds, wood etc.); further, insect herbivores are grouped by taxonomic orders. Thus, neither the network dimensions, linkage density nor effective specialization of herbivores can be compared in any way among these two major studies.
A comparison is at least feasible with several extensive studies of folivorous caterpillars sampled and reared from their host plants conducted at different latitudes in the New World (Dyer et al. 2007). In four tropical sites (Central Brazil, Panama, Costa Rica and Ecuador), 192–565 lepidopteran species were reared from 109 to 281 host species. These caterpillars seem to be more specialized than the ones in the same guild in Novotny and colleagues’ study. Still, the comparison is hampered by the fact that the surveys compiled in Dyer et al. (2007) cover different sized areas and are less standardized than the study in Papua New Guinea. More importantly, Novotny et al. show that no single guild can fully represent the gamut of herbivore associations with hosts. On different continents, some idiosyncratic combinations of guilds might radiate and diversify, whereas others could be entirely absent (Lawton, Lewinsohn & Compton 1993).
Biological research is rife with overwhelming enterprises. However, these are often amenable to technological acceleration. Mapping an entire genome, which less than two decades ago took several years to accomplish, has been shortened considerably and is fast becoming a routine task. By contrast, tracing food webs will always be a much more complex, laborious and lengthy task. Even if genetic barcoding fulfils the most optimistic hopes of a dependable and speedy sorting of most invertebrates (Janzen 2004), barcoding has to be applicable not only to adult gut contents (as in Jurado-Rivera et al. 2009) but it has to attain the effective and large-scale identification of larval hosts in the adults, to provoke a major breakthrough in trophic network mapping. Until then, the identification of trophic links will continue to depend on extensive and detailed field observations and collection, complemented by a great deal of laboratory rearing and feeding tests.
The plant–herbivore web assembled by Novotny and his many co-workers, although a patchwork quilt that is far from complete, is likely to be as good as it gets. As their synthesis paper shows, the effort is paying off by overcoming the limits of a single-guild or single-taxon study. A key conclusion is that no taxon or guild can be taken as a reliable surrogate for the full herbivore assemblage; however, a carefully chosen combination of guilds or taxa may in time fill that role. A few comparable studies strategically placed in other sites would give at least a rough idea if the patterns found in Papua New Guinea hold in tropical forests on other continents and in other ecosystems.