INTRODUCTION
Interaction networks are usually characterized by non-random topological patterns and some degree of phylogenetic signal in the interactions (Rezende et al. Reference Rezende, Lavabre, Guimarães, Jordano and Bascompte2007; Bellay et al. Reference Bellay, Lima, Takemoto and Luque2011; Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012). Different structures have been recorded and the specific configuration of an ecological network depends mainly on the type of interaction (e.g. mutualistic vs antagonistic) and the level of intimacy among species (e.g. symbiotic vs non-symbiotic interactions) (Guimarães et al. Reference Guimarães, Rico-Gray, Oliveira, Izzo, dos Reis and Thompson2007; Fontaine et al. Reference Fontaine, Guimarães, Kéfi, Loeuille, Memmott, van der Putten, van Veen and Thébault2011). Interactions involving parasites and hosts are a classical example of antagonistic network with high intimacy, and they are often characterized by a phylogenetic signal in the interactions (Fontaine et al. Reference Fontaine, Guimarães, Kéfi, Loeuille, Memmott, van der Putten, van Veen and Thébault2011; Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012; Lima et al. Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012). The network approach in studies about host–parasite interactions stand out among the traditional approaches involving only a few species, to contribute to the elucidation of the mechanisms governing these systems (Bellay et al. Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013).
The diversity of host–parasite interactions is reflected in the network structure and the similarity between parasite faunas tends to increase with host relatedness (Bellay et al. Reference Bellay, Lima, Takemoto and Luque2011; Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012; Lima et al. Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012). In addition, a convergence of ecological traits among phylogenetically distant host species may also increase the similarity among their parasite faunas (Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012). Parasite faunas may comprise species with different life strategies, which are grouped mainly as ectoparasites (with direct contact with the external environment) or endoparasites (without direct contact with the external environment) (Bush et al. Reference Bush, Fernández, Esch and Seed2001). Studies on the variations in the structure of ecto- and endoparasite interactions with hosts are still scarce.
Parasite life strategy and host phylogeny seem to affect specialization in host–parasite interactions, which in turn explains why fish–parasite networks are characterized by low levels of connectance and nestedness, and high levels of modularity (Bellay et al. Reference Bellay, Lima, Takemoto and Luque2011, Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013; Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012; Lima et al. Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012; Poulin et al. Reference Poulin, Krasnov, Pilosof and Thieltges2013). Connectance is calculated as the proportion of interactions that are actually realised in relation to the total number of interactions that could be realised in the network (Pimm, Reference Pimm1982). Nestedness occurs if the interactions of species with fewer connections in a bipartite network represent a subset of the interactions made by species with more connections (Almeida-Neto and Ulrich, Reference Almeida-Neto and Ulrich2011). On the other hand, if there are subgroups of species interacting with each other more than with other species of the same network (modules), the network has a modular structure (Mello et al. Reference Mello, Marquitti, Guimarães, Kalko, Jordano and de Aguiar2011). Although those archetypical topologies have often been considered in studies on interaction patterns in different mutualistic and antagonistic networks (see Lewinsohn and Prado, Reference Lewinsohn and Prado2006), this approach is not commonly applied to studies on host–parasite networks (Poisot et al. Reference Poisot, Stanko, Miklisová and Morand2013).
Interactions between hosts and parasites may differ among parasite groups, if their biological traits and infestation processes result in distinct interaction constraints (Poisot et al. Reference Poisot, Stanko, Miklisová and Morand2013). Therefore, by looking at similarities and differences in the structure of interactions among ecto- and endoparasites, we can gain some insight on the underlying mechanisms of host-parasite networks. Lima et al. (Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012) observed that variations in the specificity of interactions may be responsible for differences in the structure of ecto- and endoparasite-fish networks. It is known that a single individual host can harbour both ecto- and endoparasites. Some processes that may lead fish species to share ectoparasites may also lead them to share endoparasites, thereby putting those host species in the same module within a network.
Our goal in the present study was to understand the structure of networks formed by ecto- and endoparasites of fish. To fulfil this goal, we asked the following questions (i) Do fish–ectoparasite and fish–endoparasite subnetworks from the same locality differ in terms of host taxonomy and topology (proportion of parasite species per host species, connectance, nestedness and modularity)? (ii) Do host species that share the same ectoparasites also share the same endoparasites? First, we expected differences in the biology of interactions between fish and their endo- and ectoparasites to result in networks with different structures. Second, the composition of modules in fish–parasite networks was expected to reflect the taxonomic distance between host species, as host niches tend to be phylogenetically conserved (e.g. distribution in the water column, foraging strategy). Furthermore, ecto- and endoparasites infestations are different phenomena, but they are not independent from each other, because they may occur in the same individual host. Therefore, we expected high similarity in host composition between the modules found in ecto- and endoparasite networks.
MATERIALS AND METHODS
Data
Twenty-two fish–parasite networks were obtained from the literature. The number of host and parasite species in the networks range from 6 to 91 and from 20 to 420, respectively (Table 1). We built the networks as adjacency matrices with host species in the rows, parasite species in the columns and binary values in the cells (presence of absence of interaction between a i row and a j column). To control for an effect of spatial variations, we analysed pairs of subnetworks formed by either endo- or ectoparasites that belonged to the same complete network from a given locality.
Table 1. Fish–parasite networks analysed in the present study. H is the number of host species and P is the number of parasite species of each network
a See Supplementary Material (online version only).
We restricted our analysis to metazoan parasites. The studied ectoparasites belong to the following taxonomic groups: Acari, Branchiura, Copepoda, Hirudinea, Isopoda, Mollusca, Monogenea and Myxosporea. The studied endoparasites were represented by Acanthocephala, Aspidobothrea, Cestoda, Digenea, Nematoda, Pentastomida, and some species of Monogenea and Myxosporea (see Supplementary Material). In the studied networks, the larval and adult stages of a parasite species can have different niches (host species) in the same network. Therefore, different stages were regarded as different ‘functional species’ in the network, as in Vázquez et al. (Reference Vázquez, Poulin, Krasnov and Shenbrot2005) and Bellay et al. (Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013).
Network characteristics
To test for an influence of host taxonomy on host–parasite interactions, we calculated a correlation between the matrix of taxonomic distances (a proxy for phylogenetic distance; Koehler et al. Reference Koehler, Brown, Poulin, Thieltges and Fredensborg2012) between fish species and the dissimilarity matrix of parasite fauna composition with a Mantel test, using 1000 randomizations and the Pearson method in the package vegan (Oksanen et al. Reference Oksanen, Blanchet, Kindt, Legendre, Minchin, O'Hara, Simpson, Solymos, Stevens and Wagner2014) for R 3·1·1 (R Development Core Team, 2014). To build the dissimilarity matrix used in this analysis, we used the Jaccard index available in the function vegdist in the package vegan. We calculated the matrix of taxonomic distance (MTD) for each network using the following equation:
$${\rm MTD} = Md-(Cw + Ow + Fw + Gw + Sw)$$
where Md is the maximum distance found in the fish community (maximum distance = 5, referring to the taxonomic category class) and Cw, Ow, Fw, Gw and Sw are matrices for each taxonomic category (class, order, family, genus and species, respectively) generated by the function weight.taxo available in package ape (Paradis et al. Reference Paradis, Claude and Strimmer2004) for R. Nomenclature followed the taxonomic descriptions provided by FishBase (Froese and Pauly, Reference Froese and Pauly2013). Therefore, species of the same genus exhibit a value of taxonomic distance (td) equal to 1, different genera have td = 2, different families have td = 3, different orders have td = 4 and different classes have td = 5 (see Rezende et al. Reference Rezende, Lavabre, Guimarães, Jordano and Bascompte2007).
We evaluated three general descriptors of network structure: connectance (C), nestedness (NODF), and modularity (M). To control the intrinsic negative relationship between connectance and species richness (Thébault and Fontaine, Reference Thébault and Fontaine2008), we used the residual connectance instead of absolute connectance values. The residual connectance is calculated by the residuals of the simple linear regression between the log10-transformed values of observed and possible interactions in each network (e.g. Fonseca and John, Reference Fonseca and John1996). This analysis was carried out in Statistica 7·0 (Statsoft, 2005).
The degree of nestedness was calculated using the NODF index (nestedness metric based on overlap and decreasing fill; Almeida-Neto et al. Reference Almeida-Neto, Guimarães, Guimarães, Loyola and Ulrich2008). The significance of the observed NODF-values was estimated with a Monte Carlo procedure (1000 randomizations) based on the row–column probability null model, Ce, in the program Aninhado (Guimarães and Guimarães, Reference Guimarães and Guimarães2006).
To test for a modular structure in the host–parasite networks, we used a simulated annealing algorithm to calculate the degree of modularity (M) of each network (Guimerà and Amaral, Reference Guimerà and Amaral2005). Values of M = 0 indicate the absence of subgroups in the network, whereas values near the maximum (M = 1) indicate networks strongly divided into subgroups. Modularity was calculated in the program NETCARTO (Guimerà and Amaral, Reference Guimerà and Amaral2005). As NETCARTO does not include the Ce model for the estimation of significance, we used a function for R (developed by Professor Nadson RS da Silva) to estimate the significance of M. With this function, we generated 1000 randomizations of each network based on the null model Ce. For each matrix, M was calculated in NETCARTO using a Fortran code (developed by Flávia M. D. Marquitti and first used by Mello et al. Reference Mello, Marquitti, Guimarães, Kalko, Jordano and de Aguiar2011) to automate the calculation and compilation of M-values. For each network, the significance (P) was obtained from the number of random matrices with M-values equal or higher than the observed M-value, divided by the number of randomized matrices. The R scripts are available from the authors upon request.
Data analysis
Differences in the proportion of parasites per host, host taxonomy (Mantel r coefficient), residual connectance, nestedness and modularity between ecto- and endoparasite networks were tested with a Wilcoxon test for paired samples. We applied a chi-squared test to compare the frequency of significant nested and modular structure between ecto- and endoparasite networks.
If the subnetworks of ecto- and endoparasites from the same locality were significantly modular, we evaluated the similarity in the formation of modules with a Mantel test (with the same procedure mentioned above), considering only the host species that were present in both networks. For this purpose, we identified the host species in each module of the network using the program NETCARTO, and built matrices whose rows and columns corresponded to the host species present in both networks. The value ‘1’ was given to pairs of host species that occurred in the same module, and the value ‘0’ was given to pairs of host species that did not occur in the same module.
RESULTS
The species richness of ecto- and endoparasites varied in subnetworks from 6 to 181 and from 11 to 239, respectively. The values of all network descriptors obtained for each subnetwork are presented in Table 2. The endoparasite subnetworks showed a higher proportion of parasite species per host species (PPHecto: mean = 1·49; PPHendo: mean = 2·48; Wilcoxon T = 9; Z = 3·81; P < 0·001; Fig. 1a).
Fig. 1. Parameters calculated for host–parasite networks in 22 localities considering ecto- and endoparasites in separate subnetworks. (a) Proportion of parasite species per host species; (b) Mantel r statistic (the influence of host taxonomy on host–parasite interactions); (c) residual connectance; (d) nestedness and (e) modularity.
Table 2. Parameters calculated for host–parasite networks in 22 localities considering ecto- and endoparasites in separate subnetworks
Abbreviations: S, species richness; H, host species; Pa, parasite species; I, host–parasite interactions; PHP, proportion of parasite species per host species; Mr, Mantel r statistic obtained between the host taxonomic distance matrix and the host–parasite dissimilarity matrix; C, connectance; rC, residual connectance; NODF, nestedness; M, modularity; mo, module number; ecto, ectoparasite–host network; endo, endoparasite–host network.
a The identity of networks by numbers corresponds to that in Table 1.
Thirty-three (75%) out of 44 subnetworks presented a positive and significant relation of parasite fauna composition with host taxonomy. There were no differences in the Mr-values between ecto- and endoparasite subnetworks (Mrecto: mean = 0·41; Mrendo: mean = 0·46; Wilcoxon T = 46; Z = 0·40; P = 0·683; Fig. 1b). We found significant differences in connectance, nestedness and modularity between ecto- and endoparasite subnetworks. Residual connectance was higher in endoparasite subnetworks (Cr ecto: mean = 12·53; Cr endo: mean = 14·58; Wilcoxon T = 0; Z = 4·10; P < 0·001; Fig. 1c).
Endoparasite subnetworks were more nested than ectoparasite subnetworks (NODFecto: mean = 16·23; NODFendo: mean = 23·11; Wilcoxon T = 39; Z = 2·84; P = 0·004; Fig. 1d). In addition, nestedness was significant in 17 (39%) out of 44 networks, and the number of significantly nested subnetworks was higher among endoparasites (χ 2 = 4·69; gl = 1; P = 0·03).
The ectoparasite subnetworks were more modular than the endoparasite networks (M ecto: mean = 0·62; M endo: mean = 0·47; Wilcoxon T = 11; Z = 3·74; P < 0·001; Fig. 1E). Twenty-eight (64%) out of 44 networks showed significant modularity, and the frequency of the significantly modular structures did not differ between subnetwork types (χ 2 = 0; gl = 1; P = 1).
In 12 (55%) of the studied localities, both subnetworks (ecto- and endoparasites) were significantly modular. In eight of the localities with both modular subnetworks (67%) host module composition was correlated between ecto- and endoparasite subnetworks. However, we observed low Mr-values, which suggests a weak relationship (r = 0·23 ± 0·08) (Table 3).
Table 3. Mantel r statistic (Mr) calculated for the correlation between the dissimilarity in host species composition of the modules found in ecto- and endoparasites networks
a See Supplementary Material (online version only)
DISCUSSION
In the present study, we found that ecto- and endoparasite subnetworks from the same local assemblage differed in their topologies, thus implying that differences in the biology of parasitic interactions may lead to different interaction patterns at the community level. Those differences were observed in all topological metrics analysed.
One key point to consider is that the interior of host can offer a higher diversity of sites (organs and tissues) for parasite attachment than the external surface of host. This might explain, for instance, the greater number of endoparasite than ectoparasite species found. Another factor that potentially influenced the richness patterns is the various routes of infection that are available to fish endoparasites (i.e. active penetration through the skin or trophic transmission). Those routes of infection contribute to species diversity, because they increase the probability of host–parasite encounters and may reduce competition among parasite species (Poulin, Reference Poulin1998; Dobson et al. Reference Dobson, Lafferty, Kuris, Hechinger and Jetz2008; Lima et al. Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012).
Host characteristics, such as density, body size, diet and biogeographic distribution, may directly influence parasite diversity (Takemoto et al. Reference Takemoto, Pavanelli, Lizama, Luque and Poulin2005; Poulin and Leung, Reference Poulin and Leung2011; Timi et al. Reference Timi, Rossin, Alarcos, Braicovich, Cantatore and Lanfranchi2011). Host species that are phylogenetically close tend to present more similar parasite faunas than unrelated host species (Bellay et al. Reference Bellay, Lima, Takemoto and Luque2011, Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013; Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012; Lima et al. Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012). This tendency would result from parasite species persistence after speciation events of the ancestral host and of the ecological similarity of these hosts (Poulin, Reference Poulin1998), and how we observed, independent of the habitat type used by parasites (ecto- or endoparasites).
The residual connectance values were higher for endoparasite than ectoparasite networks. In networks, connectance provides important information that may allow the understanding of other structural parameters, for example, an increase in connectance can reduce the possibility of nested and modular structures simultaneously in a network (Fortuna et al. Reference Fortuna, Stouffer, Olesen, Jordano, Mouillot, Krasnov, Poulin and Bascompte2010). Due to the high specificity of host–parasite networks, the connectance values are generally low (Bellay et al. Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013). In addition, several studies have shown that the range of host species of the endoparasites of fish may be wider than the range of host species of the ectoparasites (particularly monogeneans; Strona et al. Reference Strona, Galli and Fattorini2013). The presence of endoparasites in larval stages, that tend to be more generalist than adults (Bellay et al. Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013), also contributed to increase the connectance.
In the present study, the nested structure of some networks was more closely related to the life strategy of endoparasites, which suggests differences in the organization of host–parasite networks as a function of host type (e.g. taxonomic group; aquatic or terrestrial). Studies on terrestrial hosts showed that the endoparasites have a greater degree of host specificity than the ectoparasites in the networks (see Brito et al. Reference Brito, Corso, Almeida, Ferreira, Almeida, Anjos, Mesquita and Vasconcellos2014). A high degree of specificity in the use of host species may result in relatively low levels of nestedness. This has been used as a basis to infer that mutualistic and antagonistic networks have similar organizations, particularly for a model system in which ectoparasites use terrestrial hosts (Graham et al. Reference Graham, Hassan, Burkett-Cadena, Guyer and Unnasch2009). Several hypotheses have been presented in previous studies to explain the nestedness structure in networks (see Suweis et al. Reference Suweis, Simini, Banavar and Maritan2013). For example, the ecology and factors related to the parasite life cycle may contribute to the nestedness pattern, particularly among endoparasites (Lima et al. Reference Lima, Giacomini, Takemoto, Agostinho and Bini2012). The reason for this effect is that the larval stages of these parasites tend to be more generalist than the adult parasites (Bellay et al. Reference Bellay, de Oliveira, Almeida-Neto, Lima Junior, Takemoto and Luque2013), as mentioned above. Furthermore, the adult stages may have been obtained by trophic transmission, and the host species may have nested diets, thus allowing the formation of nested parasitic fauna.
We observed no differences between subnetworks in the frequency of a significantly modular structure, but they differed from one another in their degree of modularity. The presence of specialized interactions is an important factor when interpreting the modular structure of ecological networks (Mello et al. Reference Mello, Marquitti, Guimarães, Kalko, Jordano and de Aguiar2011). Parasitism in general is expected to be highly specialized (Thompson, Reference Thompson1994), which may explain the lack of difference in the frequency of modular structures. But variations in specificity made endoparasite subnetworks be more nested than modular, while the opposite was observed for ectoparasites. The studied ectoparasite subnetworks presented an average high modularity values, probably because ectoparasites, particularly monogeneans, are more specialized than endoparasites (Strona et al. Reference Strona, Galli and Fattorini2013).
Although the co-occurrence of ecto- and endoparasites in the parasitic fauna of hosts is common, we observed a weak relationship (reflected in low Mr coefficients) between the host species composition of modules found in ecto- and endoparasite subnetworks. Thus, for example, hosts that shared the same module, when only ectoparasites were evaluated, normally tended to occur in distinct modules when we evaluated only endoparasites. Consequently, we may infer that different factors influence module organization in ecto- and endoparasite networks. However, the high specificity presented by the ectoparasites suggests that host phylogeny is a key factor in module organization (see Krasnov et al. Reference Krasnov, Fortuna, Mouillot, Khokhlova, Shenbrot and Poulin2012), whereas host diet could have a stronger influence for endoparasites than phylogeny, due to the trophic transmission of endoparasites (see Garrido-Olvera et al. Reference Garrido-Olvera, Arita and Pérez-Ponce De León2012). Future studies that include information of the phylogenetic, biological and ecological characteristics of host species may clarify which factors are most important to build up the modular structure of host–parasite networks.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit http://dx.doi.org /10.1017/S0031182015000128
ACKNOWLEDGEMENTS
The authors thank Nadson R. S. da Silva for the elaboration of the function that allowed us to run the null model Ce in R, Flávia M. D. Marquitti for kindly providing us with the Fortran code for the automation of the modularity analysis, and Vanessa M. Algarte for her suggestions to an early version of the manuscript.
FINANCIAL SUPPORT
Manuscript funded by PEA/CAPES/PROEX, DIRPPG/UTFPR – Campus Londrina and the Brazilian Research Council (CNPq). MAR Mello was sponsored by Ulm University, Humboldt Foundation (AvH, 1134644), São Paulo Research Foundation (FAPESP, 06/00265-0, 05/00587-5; 07/50633-9), Federal University of Minas Gerais (UFMG, PRPq 01/2013, 14/2013, 02/2014), Minas Gerais Research Foundation (FAPEMIG, APQ-01043-13), CNPq (472372/2013-0), Research Program on Atlantic Forest Biodiversity (PPBio-MA/CNPq) and Ecotone Inc. (‘Do Science and Get Support Program’). MAN received research fellowships (306843/2012-9 and 306870/2012-6, respectively) from CNPq. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
