Multiscale determinants of parasite abundance: A quantitative hierarchical approach for coral reef fishes

https://doi.org/10.1016/j.ijpara.2009.09.010Get rights and content

Abstract

During recent decades, there have been numerous attempts to identify the key determinants of parasite communities and several influential variables have been clarified at either infra-, component or compound community scales. However, in view of the possible complexity of interactions among determinants, the commonly-used exploratory and statistical modelling techniques have often failed to find meaningful ecological patterns from such data. Moreover, quantitative assessments of factors structuring species richness, abundance, community structure and species associations in parasite communities remain elusive. Recently, because they are ideally suited for the analysis of complex and highly interactive data, there has been increasing interest in the use of classification and regression tree analyses in several ecological fields. To date, such approaches have never been used by parasitologists for field data. This study aims to both introduce and illustrate the use of multivariate regression trees in order to investigate the determinants of parasite abundance in a multi-scale quantitative context. To do this, we used new field epidemiological data from 1489 coral reef fishes collected around two islands in French Polynesia. We evaluated the relative effect and interactions of several host traits and environmental factors on the abundance of metazoan parasite assemblage at several scales and assessed the impact of major factors on each parasite taxon. Our results suggest that the islands sampled, the host species and host size are equal predictors of parasite abundance at a global scale, whereas other factors proved to be significant predictors of a local pattern, depending on host family. We also discuss the potential use of regression trees for parasitologists as both an explorative and a promising predictive tool.

Introduction

The importance of being able to successfully model relationships between parasite species abundance and host/environmental variables has long been recognized and remains a fundamental problem in parasitology. Numerous theoretical models have been proposed to explain observed species abundance distributions and epidemiological theories point to several environmental factors and host traits that may affect abundances of macroparasite populations (Anderson and May, 1978). During the last few decades, there have been numerous attempts to identify the key determinants of parasite communities in a variety of habitats and several key variables have been elucidated at either infra-, component or compound community scales (Bell and Burt, 1991, Sousa, 1994, Poulin, 1995, Poulin, 1997, Poulin, 2004, Rohde et al., 1995, Gregory et al., 1996, Sasal et al., 1997, Thomas, 2002, Sikkel et al., 2009). However, despite these efforts, multi-scale quantitative assessment of factors structuring species richness, abundance, community structure and species associations in parasite communities have been elusive (Holmes, 1990). Only a few studies have indeed attempted to quantify the factors influencing parasite communities (Sasal et al., 1999). Defining and quantifying key factors is equally important in determining the effect of scale on parasite communities and processes, and ensuring that predictability in parasite communities is considered (Carney and Dick, 2000). It is, therefore, difficult to assess the relative importance of different host traits and environmental characteristics for the evolution of parasite diversity in general.

Species–environment relationships can be modelled using either individual species or assemblages of species. For individual species analyses, species are independently related to the environmental variables, whereas for community analyses, species are jointly related to the environment using a single model. Thus, community analyses including infra-, component and compound communities are more constrained but much more informative. Many host/environmental variables have been identified as being correlated with parasite species abundance but the relationships generally exhibit high variance. For example, many suitable hosts do not harbour a species due to environmental constraints. Because of this variability, most of the previous analytical techniques have focused on indirect ordinal methods. Ordination methods are one approach to the problem and are designed to arrange individual hosts in a theoretical space so that nearby samples in the space have similar species composition. Associations between the ordination results and environmental factors may then be explored by correlation or regression. Popular methods used for species abundance analysis include principal components analysis (PCA), canonical correspondence analysis (CCA), generalized linear models (GLM), generalized additive models (GAM), multi-dimensional scaling (MDS) and various clustering methods. Traditional ordination methods are widely used for community analyses (Sasal et al., 1997, Sasal et al., 1999, Thomas, 2002, Šimková et al., 2008, Kennedy, 2009) and perform efficiently for well designed parasitological experiments or sampling protocols. However, in view of the possible complexity of interactions among determinants, such exploratory and predictive approaches often fail to find meaningful ecological patterns from complex data. Popular methods are indeed cumbersome to use, sub-optimal or of limited utility in investigating large field epidemiological data sets (including the one analyzed in this study) (De’ath and Fabricius, 2000; Lewis, 2000. An introduction to classification and regression tree (CART) analysis. In: Annual Meeting of the Society for Academic Emergency Medicine. San Francisco, California; De’ath, 2002). There are a number of reasons for these difficulties. First, there are often many possible “predictors” and it is difficult to choose which one to use. Second, the predictors are rarely normally distributed with obvious differing degrees of variation or variance. Third, complex interactions or patterns may exist in the data. These interactions are generally difficult to model and virtually impossible to model when the number of interactions and variables becomes substantial. Co-linearity in the predictors is also a crucial problem as two highly correlated predictors may both appear non-significant even though they explain a significant proportion of the deviance if considered individually. Fourth, the results of traditional methods may be difficult to use in a predictive manner. Environmental variables can be used either directly or indirectly to explain the variability in parasite abundance. When used directly, the species and environmental data jointly determine the variability. When used indirectly, the structure of the species data is first determined and is then related to the environmental data. In most cases, traditional ordination methods neglect direct analysis with very few attempts to investigate direct analytical methods.

Within the last 10 years, there has been increasing interest in the use of classification and regression tree (CART) analysis for complex ecological data that may include lack of balance, outliers, missing values, non-linear relationships between variables, and high-order interactions (De’ath and Fabricius, 2000). This is important as the abundance of parasites may be determined by a large number of interactive factors. Within CART, multivariate regression trees (MRT) are relatively new in ecology (De’ath, 2002) and are virtually unexplored for complex epidemiological field data with multiple response variables and multiple explanatory variables. To date, CARTs have been used rarely by parasitologists for field data (but see Vignon et al., in press).

Among vertebrate hosts, the abundance and richness of the parasite fauna of fish (both marine and freshwater) have been extensively studied (Poulin, 2004). They have contributed considerably to the development of our general understanding of parasitological determinants in aquatic ecosystems. Although parasite diversity in coral reefs has not been thoroughly evaluated (Morand et al., 2000, Muñoz et al., 2007), data generally show that parasite biodiversity in these ecosystems is very high, reaching more than two to 10 times the number of fish species (Rohde, 1976, Lester and Sewell, 1989, Cribb et al., 1994).

This study aims to both introduce and illustrate the use of MRT in order to investigate the determinants of metazoan parasites abundance in a multi-scale quantitative context. To do this, we used new field epidemiological data from 1489 coral reef fishes belonging to two non-related families (Lutjanidae and Serranidae) collected off two Polynesian Islands. Only general trends are presented for each scale analysed. MRT gives us the opportunity to analyse the abundance of the parasite communities at multiple scales and to quantify the relative importance of each factor tested, as well as the impact of major factors on each parasitic taxon. We finally discuss the potential use of MRT by parasitologists as both an explorative and predictive tool.

Section snippets

Sites, host and parasite collection

A total of 1489 coral reef fish representing eight species were collected between 2005 and 2007 during an extensive survey of macro-parasites from two islands in the South Pacific Ocean (Moorea Island, 17°30’S, 149°50’W, Society archipelago, French Polynesia and Ua Huka, 8°57’S, 139°35’W, Marquesas Islands, French Polynesia) (Table 1). Fish were collected using a spear gun from both the inner and outer reefs at depths ranging from 5 to 40 m. Immediately after collection, each fish was

Pattern in parasite abundances

The abundance of parasites sharply contrasted among host individuals. There was on average 26 ± 54 (SD) parasites per fish, belonging to 3 ± 2 parasite species. Fig. 1 shows a composite exploratory tree, allowing overall visual interpretation of both relative importance and interactions among factors at multi-scales. This composite tree provides strong evidence of the complexity of epidemiological data and the synergy (high-order interactions) among several factors. For trees with a single split,

Discussion

Studying parasite diversity and the determinants of this diversity have been one of the major subjects of meta-analyses conducted during the last 10 years in parasite ecology. However, most of the studies considered hosts from different origins, potentially inducing confounding effects because parasites are not equally distributed in the host sampling area. Here, we provided strong evidence that several factors determine parasite abundance, with a prominent role for islands, host species,

Acknowledgements

We are grateful to and thank Dr. Rod Bray for revising this manuscript. The authors thank anonymous reviewers who made valuable suggestions to improve the manuscript.

References (53)

  • L. Breiman et al.

    Classification and Regression Trees

    (1984)
  • A.O. Bush et al.

    Ecological versus phylogenetic determinants of helminth parasite community richness

    Evol. Ecol.

    (1990)
  • J.P. Carney et al.

    Helminth communities of yellow perch (Perca flavescens (Mitchill)): determinants of pattern

    Can. J. Zool.

    (2000)
  • T.H. Cribb et al.

    Ecology and diversity of digenean trematodes of fishes of the Great Barrier Reef

    Int. J. Parasitol.

    (1994)
  • G. De’ath

    Multivariate regression trees: A new technique for modeling species-environment relationships

    Ecology

    (2002)
  • G. De’ath et al.

    Classification and regression trees: a powerful yet simple technique for ecological data analysis

    Ecology

    (2000)
  • M. Dufrêne et al.

    Species assemblages and indicator species: the need for a flexible asymmetrical approach

    Ecol. Monogr.

    (1997)
  • J. Elith et al.

    Novel methods improve prediction of species’ distributions from occurrence data

    Ecography

    (2006)
  • J. Elith et al.

    A working guide to boosted regression trees

    J. Anim. Ecol.

    (2008)
  • J.H. Friedman

    Greedy function approximation: a gradient boosting machine

    Ann. Stat.

    (2001)
  • J.H. Friedman et al.

    Additive logistic regression: a statistical view of boosting

    Ann. Stat.

    (2000)
  • R.D. Gregory et al.

    Helminth parasite richness among vertebrates

    Biodivers. Conserv.

    (1996)
  • A.S. Grutter

    Comparison of methods for sampling ectoparasites from coral reef fishes

    Mar. Freshwater Res.

    (1995)
  • J.F. Guégan et al.

    A nested parasite species subset pattern in tropical fish: host as major determinant of parasite infracommunity structure

    Oecologia

    (1994)
  • J.F. Guégan et al.

    Can host body size explain the parasite species richness in tropical freshwater fishes?

    Oecologia

    (1992)
  • J.C. Holmes

    Helminth communities in marine fishes

  • Cited by (20)

    • Butterflyfish gill mucus metabolome reflects diet preferences and gill parasite intensities

      2022, Applied Environmental Metabolomics: Community Insights and Guidance from the Field
    • Patterns of monogenean abundance in native and invasive populations of Planiliza haematocheila (Teleostei: Mugilidae): interactions between climate and host defence mechanisms explain parasite release

      2020, International Journal for Parasitology
      Citation Excerpt :

      Ligophorus spp. infect the so-iuy mullet beginning from the second month of its life and last until death with increasing abundance (Sarabeev and Domnich, 2000 and the present study). Changes in parasite abundance and intensity related to fish size are well known and broadly discussed in the literature (Knoff et al., 1997; Poulin, 2000; Bagge et al., 2004; Luque et al., 2008; Vignon and Sasal, 2010; Amarante et al., 2016). A positive correlation between host body length and abundance of ectoparasites is based on increased resource availability, since larger fish provide more attachment space and can thus sustain more parasite individuals (Bagge et al., 2004; Amarante et al., 2016).

    • Drivers of parasite community structure in fishes of the continental shelf of the Western Mediterranean: the importance of host phylogeny and autecological traits

      2019, International Journal for Parasitology
      Citation Excerpt :

      Considering parasites in host assemblages rather than in a single host species can provide a better understanding of the factors driving the structure and composition of parasite communities, and broaden the possibilities associated with their use as biological and environmental indicators. A number of studies have been devoted to exploring parasite community patterns in marine and freshwater fish assemblages (e.g. Takemoto et al., 2005; Luque and Poulin, 2008; Valtonen et al., 2010; Vignon and Sasal, 2010; Bellay et al., 2011; Poulin and Leung, 2011; Lima et al., 2016; Baia et al., 2018; Calhoun et al., 2018). These studies have highlighted the prominent role of host phylogeny and/or ecological convergence, in terms of similar habitat use or trophic level, in explaining the resemblance among fish parasite communities (Marcogliese, 2002; Luque and Poulin, 2008).

    • Did biogeographical processes shape the monogenean community of butterflyfishes in the tropical Indo-west Pacific region?

      2017, International Journal for Parasitology
      Citation Excerpt :

      In a recent study, Locke et al., 2012 showed that distance decay depends upon the geographic scale of the study and latitude, being more significant further from the Equator. However, Vignon and Sasal (2009) found that fish parasite abundances at a global scale were determined by the site sampled, host species and host size. To date, most studies of parasite communities have used either presence/absence or abundance data to analyse variation in the overall similarity in structures of communities (e.g. Jaccard, Sørensen, Bray-Curtis algorithms).

    • Do captive conditions favor shedding of parasites in the reared Atlantic bluefin tuna (Thunnus thynnus)?

      2011, Parasitology International
      Citation Excerpt :

      In order to build a decision tree to predict classification of tuna in group B (before) or A (after) based on didymozoids abundances (other parasitic groups were excluded) and year of sampling, DTREG software was used. Detailed logic behind this prediction tool was explained recently by Vignon and Sasal [38]. Target variable as the one whose value needs to be modeled and predicted as a function of the predictor variables was chosen to be “group” (B or A, respectively).

    View all citing articles on Scopus
    View full text