To Be or Not to Be Associated: Power study of four statistical modeling approaches to identify parasite associations in cross-sectional studies

ORIGINAL RESEARCH ARTICLE published: 15 May 2014 doi: 10.3389/fcimb.2014.00062 CELLULAR AND INFECTION MICROBIOLOGY To be or not to be associated: power study of four statistical modeling approaches to identify parasite associations in cross-sectional studies Elise Vaumourin 1,2*, Gwenaël Vourc’h 1 , Sandra Telfer 3 , Xavier Lambin 3 , Diaeldin Salih 4 , Ulrike Seitzer 5 , Serge Morand 6,7 , Nathalie Charbonnel 8 , Muriel Vayssier-Taussat 2 and Patrick Gasqui 1 1 INRA, UR346 Epidémiologie Animale, Saint Genès Champanelle, France INRA-Anses-ENVA, USC BIPAR, Maisons-Alfort, France 3 School of Biological Sciences, University of Aberdeen, Aberdeen, UK 4 Department of Ticks and Tick-borne Diseases, Veterinary Research Institute, Khartoum, Sudan 5 Division of Veterinary-Infection Biology and Immunology, Research Center Borstel, Borstel, Germany 6 Institut des Sciences de l’Evolution (CNRS /IRD / UM2), University of Montpellier 2, Montpellier, France 7 Animal et Gestion Intégrée des Risques, CIRAD, Montpellier, France 8 INRA, UMR CBGP (INRA / IRD / CIRAD / Montpellier SupAgro), Montpellier, France 2 Edited by: Jean François Cosson, National Institute of Agricultural Research, France Reviewed by: Valerio Iebba, ’Sapienza’ University of Rome, Italy Bordes Frederic, Institut des Sciences de L’Evolution (ISEM) CNRS, France *Correspondence: Elise Vaumourin, UR346 Epidémiologie Animale, INRA, Centre de Recherche de Clermont-Ferrand – Theix, Route de Theix, 63122 Saint Genès Champanelle, France e-mail: elise.vaumourin@ clermont.inra.fr A growing number of studies are reporting simultaneous infections by parasites in many different hosts. The detection of whether these parasites are significantly associated is important in medicine and epidemiology. Numerous approaches to detect associations are available, but only a few provide statistical tests. Furthermore, they generally test for an overall detection of association and do not identify which parasite is associated with which other one. Here, we developed a new approach, the association screening approach, to detect the overall and the detail of multi-parasite associations. We studied the power of this new approach and of three other known ones (i.e., the generalized chi-square, the network and the multinomial GLM approaches) to identify parasite associations either due to parasite interactions or to confounding factors. We applied these four approaches to detect associations within two populations of multi-infected hosts: (1) rodents infected with Bartonella sp., Babesia microti and Anaplasma phagocytophilum and (2) bovine population infected with Theileria sp. and Babesia sp. We found that the best power is obtained with the screening model and the generalized chi-square test. The differentiation between associations, which are due to confounding factors and parasite interactions was not possible. The screening approach significantly identified associations between Bartonella doshiae and B. microti, and between T. parva, T. mutans, and T. velifera. Thus, the screening approach was relevant to test the overall presence of parasite associations and identify the parasite combinations that are significantly over- or under-represented. Unraveling whether the associations are due to real biological interactions or confounding factors should be further investigated. Nevertheless, in the age of genomics and the advent of new technologies, it is a considerable asset to speed up researches focusing on the mechanisms driving interactions between parasites. Keywords: associations, interactions, modeling, parasite community, screening, GLM approach, network model, chi-square test INTRODUCTION A growing number of studies of many mammal hosts, including wild and domestic animals and humans, are reporting simultaneous infections by different microparasites (Cox, 2001; Palacios et al., 2009; Saisongkorh et al., 2009; Tadin et al., 2012; Jacquot et al., 2014), macroparasites (Byrne et al., 2003; Behnke, 2009; Fenton et al., 2010) and both (Jolles et al., 2008; Ezenwa and Jolles, 2011; Nunn et al., 2014). The frequency of co-occurrence can be influenced by interactions between parasites. These interactions are of crucial medical concern because they can alter host susceptibility, infection length and clinical symptoms, as illustrated by the influence of helminths on malaria severity (Nacher, 2002). From an epidemiological point of view, interactions can alter the Frontiers in Cellular and Infection Microbiology risk of transmission. Parasites can interact in a synergistic manner when the presence of one favors the infection by a subsequent parasite, as, for example, HIV and Mycobacterium tuberculosis (Corbett et al., 2003). Parasites can also interact in an antagonistic manner, as, for example, in Aedes aegypti mosquitoes, where infection with the symbiotic Wolbachia prevents subsequent infection with dengue virus, Chikungunya virus and the agent of malaria (Moreira et al., 2009). Parasite interactions have mostly been considered as a one-to-one interaction, where the infection of one parasite influences the acquisition of and/or dynamics of infection by a second parasite. However, interactions between a set of parasites are conceivable where different parasites interact within a network or through “cascade consequence” www.frontiersin.org May 2014 | Volume 4 | Article 62 | 1 Vaumourin et al. Modeling parasite associations (Rigaud et al., 2010; Bordes and Morand, 2011). For instance, such networks have been successfully used to identify interactions in ecology, e.g., El Niño (Trenberth and Fasullo, 2011), in genetics, e.g., HLA genes (Wansen et al., 1997), or in metabolic pathways, e.g., metabolic regulation (Matsuoka and Shimizu, 2012). The co-occurrence of parasites can also result from confounding factors that create statistical associations between parasites, even though there are no true biological interactions. For instance, similarities in host environment, behavior or susceptibility can cause correlations in the risk of infection between two parasites (e.g., association filters, Combes, 2001). For example, associations in humans between the agent of malaria and helminth infections may be due, in certain contexts, to common social or environmental factors, which can be depicted by a social network analysis, rather than a true biological interaction (Mwangi et al., 2006). Thus, in host populations, interactions between two parasites are suspected when the probability of coinfection is not random once confounding factors have been taken into account. In populational studies, longitudinal or time series data are useful for identifying parasite associations. However, such studies are resource-intensive. In such studies, one can test whether the presence of a parasite impacts the probability of infection by another one (e.g., Mahiane et al., 2010; Sherlock et al., 2013) or one can test whether the infection dynamics of several paras (...truncated)