Ecological genetics of invasive alien species

Abstract

There is growing realisation that integrating genetics and ecology is critical in the context of biological invasions, since the two are explicitly linked. So far, the focus of ecological genetics of invasive alien species (IAS) has been on determining the sources and routes of invasions, and the genetic make-up of founding populations, which is critical for defining and testing ecological and evolutionary hypotheses. However an ecological genetics approach can be extended to investigate questions about invasion success and impacts on native, recipient species. Here, we discuss recent progress in the field, provide overviews of recent methodological advances, and highlight areas that we believe are of particular interest for future research. First, we discuss the main insights from studies that have inferred source populations and invasion routes using molecular genetic data, with particular focus on the role of genetic diversity, adaptation and admixture in invasion success. Second, we consider how genetic tools can lead to a better understanding of patterns of dispersal, which is critical to predicting the spread of invasive species, and how studying invasions can shed light on the evolution of dispersal. Finally, we explore the potential for combining molecular genetic data and ecological network modelling to investigate community interactions such as those between predator and prey, and host and parasite. We conclude that invasions are excellent model systems for understanding the role of natural selection in shaping phenotypes and that an ecological genetics approach offers great potential for addressing fundamental questions in invasion biology.

Appendices

Appendix 1: Introduction to molecular genetic methods for inferring source populations and invasion routes

Two types of method have been used to make inferences concerning source populations and invasion routes: direct methods based on current and historical observations of IAS and indirect methods based on patterns in molecular data.

Direct methods are based on records of the presence and absence of invasive taxa. Routine controls carried out in airports and harbours by quarantine services and monitoring by environmental or agricultural agencies are particularly informative in this respect (Work et al. 2005). However, it is rarely possible to infer the routes of invasion with a high degree of precision by these direct methods. Indeed, given the low rates of establishment and expansion recorded for introduced individuals (Williamson 2006), there is no guarantee that the individuals intercepted would have spearheaded a successful invasion.

Indirect methods are based on the genetic patterns observed within and between populations at molecular markers. Traditional statistical treatments include the construction of trees from matrices of genetic distances between populations (e.g. Lozier et al. 2009; Thibault et al. 2009), parsimony networks (e.g. Voisin et al. 2005; Hoos et al. 2010) and the calculations of assignment likelihood (e.g. Genton et al. 2005; Ciosi et al. 2008). More recently, a number of studies have used clustering methods like those implemented in STRUCTURE (Pritchard et al. 2000). If the invasive population clusters clearly with one of the potential source populations, this is considered to provide fairly conclusive information about the origin of the invasive population (e.g. Marrs et al. 2008; Rollins et al. 2009). A shared ancestry of the individuals of invading populations with various populations from the native area is sometimes interpreted as evidence for an admixture origin of the invasive population considered, although it may also reflect the presence of unsampled sources, drift, or insufficient numbers of markers (Darling et al. 2008; Rosenthal et al. 2008). It is worth stressing that, although the abovementioned indirect methods have proved useful in many cases, they are all subject to two major limitations: (1) they poorly take into account the stochasticity of the demographic and genetic history considered and (2) they do not allow probabilistic estimations of competing introduction scenarios (e.g. Knowles and Maddison 2002).

Recently, a new indirect method called approximate Bayesian computation (ABC, Beaumont et al. 2002, implemented in DIYABC, Cornuet et al. 2008) has been proposed and used to draw inferences from molecular and historical data, about the complex evolutionary scenarios typically encountered in the introduction histories of IAS (Fig. 1). General statistical features of ABC have been reviewed in two recent papers (Bertorelle et al. 2010; Csillery et al. 2010) and some practical aspects that are important when using this method to make inferences about invasion routes can be found in Estoup and Guillemaud (2010). Briefly, ABC is a model-based Bayesian approach in which the posterior probabilities of different models and/or the posterior distributions of the demographic parameters under a given model are determined by measuring the similarity between the observed data set (i.e. the target) and a large number of simulated data sets. ABC has four main advantages over the more traditional indirect methods described above: (1) it uses all the data simultaneously in inference, (2) it can be used to estimate probabilities, with confidence intervals for each of the scenarios compared (e.g. Cornuet et al. 2008, Fig. 1), (3) it allows the evaluation of the power of a given analysis on the basis of controlled simulated datasets (Cornuet et al. 2008; Guillemaud et al. 2010), and (4) it avoids the introduction of misleading biases, such as those due to unsampled populations (Guillemaud et al. 2010) or genetic admixture between multiple sources (Lombaert et al. 2010). ABC thus constitutes a real advance for inferring source populations and invasion routes.

Appendix 2: Introduction to molecular genetic methods for investigating dispersal

The rapidly developing field of landscape genetics aims to understand how population genetic processes are affected by spatial and temporal environmental heterogeneity, by integrating population genetics with landscape ecology and spatial statistics. Landscape genetics approaches enable two major insights into dispersal: first, individuals with multilocus genotypes that are representative of a population other than the one they were sampled in can be identified. This is a powerful way of identifying immigrants and therefore quantifying dispersal (e.g. Guillot et al. 2005a, b). Second, the pattern of spatial genetic structuring can be tested for correlations with landscape or environmental features, allowing identification of genetic continuity (or connectivity) between patches, or discontinuities resulting from barriers to dispersal (see e.g. Balkenhol et al. 2009; Guillot et al. 2009; Storfer et al. 2010, for recent reviews).

Under a landscape genetics approach, the individual is the unit of study, and their exact geographic location must be recorded. Populations do not have to be identified a priori. Bayesian statistics are used to assign individuals to populations according to their multilocus genotypes, using software that employ clustering algorithms based on pre-defined population genetic models (e.g. STRUCTURE, Pritchard et al. 2000, see also “Appendix 1” section). In recent years, new technologies have greatly assisted marker development, vastly increasing the amount of data that can be collected, and decreasing the computation time required for data analysis. For example, new statistical approaches such as Discriminant Analysis of Principal Components (DAPC, Jombart et al. 2010), offer great potential for assigning individuals into clusters with minimal computing time when datasets are large, and when there is low population structure.

Isolation by distance tests (e.g. Mantel test) have long been used to identify correlations between genetic distance and environmental variables, but new statistical approaches are also being developed to model the relationship between genetic structuring and the environment, which allow inferences on the microevolutionary processes generating spatial genetic structure (see e.g. Guillot et al. 2009). Geographical information systems-based landscape analysis overlays landscape variables onto population genetic data to visualise patterns of genetic structuring (for example using ArcGIS or PATHMATRIX, Ray 2005), allowing environmental parameters likely to influence dispersal in heterogenous environments to be investigated. The spatial domain occupied by inferred clusters can be examined to identify dispersal barriers, using programmes such as GENELAND (Guillot et al. 2005b), and genetic diversity can be simulated, accounting for environmental and spatial heterogeneity, using software such as SPLATCHE (Currat et al. 2004). This latter approach has been modified to reconstruct invasion scenarios, investigating parameters such as dispersal distance and speed (Estoup et al. 2010). These types of simulations show how demographic processes interact with landscape features to determine spatial genetic structure (Epperson et al. 2010) and to investigate how dispersal is affected not only by obvious geographical features (e.g. mountain ranges), but also by more subtle habitat characteristics (e.g. Davis et al. 2010). They therefore offer great potential for understanding dispersal ability, and ultimately, generating information that can be used to predict the spread of IAS.

Appendix 3: Introduction to molecular genetic methods for investigating community interactions

So far, molecular studies that have attempted to investigate the strength and structure of predator–prey and parasite-host interactions, within a community context, have primarily used standard PCR (e.g. Symondson 2002; Harper et al. 2005; Sheppard and Harwood 2005). The advantage of such markers is to be able to qualitatively evaluate specific interactions between a predator and its prey or a parasitoid and its host. On the other hand, developing species–specific molecular probes can be long and costly (see Aebi et al. 2011), and the development of species-specific markers to describe whole community’s food web structure is impractical. Advances in second and next generation sequencing offer great promise as they do not rely on design of species-specific primers, are extremely sensitive, and could be used to create quantitative interaction networks. For example, Roche/454 massively parallel pyrosequencing offers considerable scope for investigating community interactions. By generating tags from 16S or 18S rDNA, data is generated for almost every organism in a sample to reveal previously uncharacterised aspects of the biological diversity (e.g. Dethlefsen et al. 2008). Datasets can then be compared to see how they differ in terms of composition. A particular advantage to this method is that many individual samples can be tagged, pooled, and sequenced in parallel (e.g. Meyer et al. 2008), and several populations can be investigated simultaneously (by “gasketting”, i.e. splitting a 454 picotiter plate into several sections). This technique has already proven successful in assessing biological diversity in the ocean (e.g. Sogin et al. 2006; Huber et al. 2007), soil (e.g. Leininger et al. 2006), and in the human body (e.g. Dethlefsen et al. 2008). Of particular relevance, a metagenomic survey of 454 sequence data from 16S and 18S rDNA in honeybee, A. mellifera hives uncovered presence of bacteria, fungi, parasites, metazoa, and viruses and found strong correlation between a particular virus and colony collapse disorder (Cox-Foster et al. 2007). A major challenge is to block amplification of the host DNA, but this can be achieved with the use of “blocking primers” (Vestheim and Jarman 2008).

Another challenge with this type of analysis is dealing with the volume of data generated. However, since metagenomics is an established method, several bioinformatics pipeline options already exist. For example, MG-RAST is a fully-automated service for annotating metagenome samples including phylogenetic classification (Meyer et al. 2008). MEGAN (http://ab.inf.uni-tuebingen.de/software/megan/welcome.html) and CARMA (http://www.cebitec.uni-bielefeld.de/brf/carma/carma.html) are also specific for metagenomics analysis to analyse large data sets and group operational taxonomic units (OTUs). Homology detection can be performed by comparing 16S and 18S sequences to reference databases such as SILVA (http://www.arb-silva.de/) using (for example) BLAT (BLAST-like alignment tool, Kent 2002) and OTUs defined based on multiple sequence alignment (Dethlefsen et al. 2008).