The ecology of pathogens and the emergence of diseases in multi-host systems are complex[1, 2]. Understanding epidemiology often requires the incorporation of a wide variety of different kinds of evidence and disciplinary approaches. Traditional surveillance and control approaches have often focused on humans, domestic animals, and known vectors. However, an increasing body of information indicates that effective disease surveillance and control may be heavily dependent on understanding the epidemiology of pathogens in relation to the ecology of their wild hosts e.g.,[4–7].
Avian Influenza Viruses (AIVs) in wild birds have recently received increased attention due to the emergence of the Highly Pathogenic AIV H5N1 strain and its potential threat to human health. Although numerous studies of low pathogenic AIV strains (LPAI) in waterfowl and wild birds have been published, encompassing tens of thousands of sampled wild birds, we still know relatively little about the susceptibility of individual bird species to AIV in relation to the global number of bird species. The avian community in a single ecosystem can include hundreds of interacting species. In addition, the response of bird species to specific AIV subtypes (16 hemagglutinins and 9 neuramidases known) is variable and prevalence patterns of specific subtypes will be determined by the bird cenosis. So far, most studies of AIV have concentrated on anseriforms and to a lesser extent on charadriiforms, which are known to be reservoirs for LPAI[7, 9]. In their synthesis of wild bird low pathogenic avian influenza surveillance worldwide, Olsen et al. found that out of more than 90 000 birds sampled, 54% were anseriforms and 25% charadriiforms. As a consequence, little information on AIV prevalence in the rest of the avian community has been published, and much of what has been published has been obtained as “by-catch” from capture protocols that have been focused on ducks. The minimum sample sizes that would be necessary to confidently estimate prevalence for most non-target bird species are often not reached, with the risk that the common practice of focusing mainly on anseriforms may be overlooking the role of other bird groups in the epidemiology of AIV in waterfowl communities.
Biases in the selection of species to sample are not the only problem in available data sets for AIV. The comparison between sample and community composition is a fundamental parameter in epidemiological studies. In many cases, a lack of information regarding the composition of the wild bird community from which the sample is taken makes conclusions from AIV studies even harder to interpret. A total of 100 positive samples from mallard ducks (Anas platyrhynchos), for example, carries quite different epidemiological implications if mallards represent 0.1% versus 90% of the number of wild birds present in the ecosystem; and similarly, the relevance of 100 positive samples from mallards differs if the system contains 10 or 100 other species. Interpretation of the role of a species in pathogen maintenance cannot be done rigorously without considering the potential role of the rest of the community. The sampling bias that is attendant on any field captures of wild birds should therefore be a crucial parameter in wildlife epidemiological studies.
As more host species are considered in a host-pathogen system, the number of potential interactions (and hence, the complexity of potential pathogen transmission pathways) increases exponentially in relation to the number of species in the analysis. The problem is further complicated, in the case of AIV, by the existence of a diversity of viral subtypes and substantive variations in pathogen-host interactions (e.g. susceptibility and pathogenicity). It rapidly becomes both empirically and computationally unfeasible to analyse the specific relationships between each host species and each pathogen subtype; and assigning each host species to a specific role in the epidemiological cycle (e.g. reservoir, dead end-host, spreader) can be extremely difficult even when large, detailed data sets are available. Often, in such contexts, management and policy responses must nonetheless occur. The control of emerging pathogens, for instance, typically requires rapid responses that are based on partial and imperfect information. There is therefore a clear need for techniques that can be used to summarize epidemiological complexity without oversimplifying it, even if resulting conclusions are later modified by the findings of more intensive studies.
This kind of problem, in which the number of interacting elements and interactions rapidly exceeds what can reasonably be measured in a typical scientific study, is common in research on complex systems. For example, overwhelming complexity generated by multiple interacting influences is a unifying problem in research on systems as diverse as gene expression, stock markets, and ecosystems. Epidemiologists have generally responded to interaction complexity by simplifying the description of what constitutes the system, deliberately excluding potentially interacting members of the epidemiological network. While this approach has produced some successes, particularly in understanding specialist pathogens with simple transmission cycles and limited numbers of interacting hosts and vectors, it also carries some potentially serious weaknesses in both theoretical and applied realms (as highlighted by analyses of the boundary specification problem in network analysis; e.g.).
An alternative approach to system simplification (i.e. rather than selectively picking out a small subset of interacting species to consider as “the system”) comes from the field of community ecology, in which researchers have attempted for decades to deconstruct the complexity of food webs. Concepts such as trophic levels and foraging guilds have played an important role in the development of ecological theory; and Elton’s trophic pyramid, in which differences in the biomass of different trophic levels are explained by the second law of thermodynamics, is one of community ecology’s most fundamental generalisations. Many of the approaches that have been developed for food web analysis in ecology are readily applicable to the analysis of the ecology of pathogen transmission in multi-host systems[1, 2, 13, 14].
The idea underlying functional group analysis is that broad, community-level trends in processes of interest can be detected by replacing a taxonomic classification of hosts with a classification that groups hosts according to their functional role in the epidemiology of a pathogen or a group of pathogens. Although “AIV” describes a group of pathogens, we treat it as a single pathogen, ignoring AIV subtype variability, because the sparse information available does not suggest that AIV modes of transmission significantly differ among subtypes (see[16, 17] for details). Hosts in an epidemiological functional group (EFG) share a common function in the epidemiology of the pathogen(s) of interest. We used the concept of EFGs to (1) investigate the ecology of AIVs in three different wild bird communities in southern Africa; and (2) critique the current scientific paradigm for field investigations of AIV in wild birds.
We used a wild bird census dataset to first allocate species to EFGs according to two epidemiological functions (reservoir vs. non-reservoir species, and the potential to introduce AIV strains through migratory behaviour; note that the functional groups, as described later, differ from the functions themselves) according to current AIV epidemiology dogma. We ranked each group in relation to their expected contribution to pathogen prevalence and then used our rankings to calculate relative a priori risk for each group. We then compared these a priori relative risks (which are effectively predictions, generated by accepted knowledge) to our empirical data on observed prevalence per group. In other words, we used EFGs as a way of exploring the degree to which empirical data match commonly held assumptions, rather than adopting the commoner approach of attempting to classify species into EFGs based on our own data. Our results suggest that commonly held assumptions may require some re-thinking.