Back to the future: Exploring riverine macroinvertebrate communities' invasibility

Riverine communities have been subject to numerous biological invasion events, with crustaceans among the most successful group of invasive animals worldwide. Understanding what makes a river system prone to invasion is of considerable interest to environmental regulators, resource managers, scientists and wider society globally. The Ponto‐Caspian amphipod, Dikerogammarus haemobaphes (Crustacea: Gammaridae), is a hyper‐successful invasive species that was first recorded in the UK in 2012. The use of contemporary distribution data for D. haemobaphes (2009–2020) from England provided a unique opportunity to study faunal community patterns and differences between sites that experienced invasion compared to those that have not. Macroinvertebrate community taxonomic, functional and phylogenetic features, as well as the presence of co‐occurrent invaders and abiotic features of the river systems, were examined from sites before the invasion and compared to control sites that were not invaded during the study period. Sites that would later experience invasion by D. haemobaphes were characterized by higher abundances of other invaders (e.g., especially Ponto Caspian taxa), lower abundances of crustaceans and typically had greater channel width and water depth. These basic characteristics may help identify sites at risk of future invasion by D. haemobaphes. Most biomonitoring tools examined displayed no difference between control and pre‐invaded samples, while both taxonomic and functional richness displayed higher values at sites that were subsequently invaded, questioning classic biological invasion hypotheses. Recognizing specific community characteristics that may be a precondition for subsequent invasion is essential for understanding and better predicting their future trajectories of change.

studied compared to their terrestrial counterparts. This raises a series of challenges and potential consequences if early detection and management are not possible (Moorhouse & Macdonald, 2015). Given that the eradication of invaders can be particularly complicated in riverine ecosystems (Simberloff, 2020), and prevention is widely accepted to be the best economic management option (Leung et al., 2002), it is critical that we improve our ability to predict taxa invasiveness. In addition, understanding what makes a community prone to invasion and what influences its "invasibility" (i.e., the susceptibility of communities to be invaded) is of considerable scientific interest for stakeholders including environmental regulators, river managers and academic scientists.
Maps obtained through species distribution models are often used to predict the potential invasive range of introduced species and help in early detection programmes (e.g., Barbet-Massin, Rome, Villemant, & Courchamp, 2018), but are typically developed using large scale variables (e.g., climate) which lack details at the local and community-scale. The availability of long-term observational field data (both biotic and abiotic information) provides the opportunity to take an additional step by predicting the invasibility of sites by specific invasive species and develop greater understanding regarding what makes some biological communities/sites susceptible to invasion compared to others at the community and ecosystem level (e.g., Cuthbert, Kotronaki, Dick, & Briski, 2020;Mathers et al., 2020). In addition, the exploration of long-term riverine community data facilitates the testing of key ecological concepts and hypotheses associated with biological invasion that have almost exclusively been developed using terrestrial ecosystems thus far (e.g., vegetation communities as model systems, Catford, Jansson, & Nilsson, 2009;Jeschke & Heger, 2018).
In this research, we focus on Dikerogammarus haemobaphes (Eichwald, 1841) (Crustacea: Gammaridae) as a model organism, and an example of a successful invasive organism in UK rivers (Guareschi, Laini, England, Johns et al., 2021). This amphipod (hereinafter Dh) originates from the Ponto-Caspian region and was first recorded in the UK during 2012, subsequently spreading throughout rivers of central-south England (Johns, Smith, Homann, & England, 2018). D. haemobaphes displays a number of features that may promote its invasiveness, including its flexible feeding habit and high fecundity (Bacela-Spychalska & Van Der Velde, 2013). More recently it has been shown to have implications on leaf litter processing efficiency (Constable & Birkby, 2016), measures of macroinvertebrate community diversity and integrity (Guareschi, Laini, England, Johns, et al., 2021) also being potentially responsible for the introduction of pathogens within its invaded range (Bojko et al., 2018).
Using long term observational macroinvertebrate data from British rivers, we analysed the patterns of biomonitoring indices, functional, phylogenetic and biocontamination metrics, as well as abiotic information between sites with different invasion trajectories. Sites were examined before the arrival of Dh, specifically considering those sites that would be invaded in the future (locations invaded at some point during the study period) and control sites (not invaded during the study period) to investigate conditions favouring the arrival and establishment of the species. In addition, the use of long time-series allows the examination of hypotheses associated with the establishment of invasive species specifically considering lotic invertebrates within riverine ecosystems and using Dh as an example of a successful biological invasion process. For example, analysing biomonitoring indices and land use types facilitate the assessment of the "human activity" hypothesis (Jeschke & Heger, 2018;Leprieur, Beauchard, Blanchet, Oberdorff, & Brosse, 2008) and analysis of the cooccurrence of other invaders permits the examination of the "invasional meltdown" hypothesis (Simberloff & Von Holle, 1999, see Table 1).
In this study we aim (a) to identify specific biotic and abiotic conditions that make riverine ecosystems susceptible to Dh invasion; and (b) to explore the validity of some of the most popular concepts and hypotheses associated with biological invasions on riverine communities. Identifying specific community and environmental characteristics that may precondition them to subsequent successful invasion is essential for understanding their future trajectories of change and may help resource managers and agencies to manage and reduce the potential for future invasions.

| Macroinvertebrate data and dataset building
Biological data were obtained from the Environment Agency, the statutory regulator within England, who is responsible for monitoring the ecological quality of rivers, lakes and coastal waterbodies. All samples/sites used had not been subject to any other known disturbance events (e.g., pollution incidents) and had not been invaded by other Dikerogammarus species. All benthic invertebrate samples were collected using the same sampling protocol for river biomonitoring in the UK, comprising a 3-minute "kicksample" using a standard pond-net (ISO 7828-1985) covering all available habitats in proportion to their occurrence, followed by a 1-min hand search (RIVPACS Macroinvertebrate Sampling Protocol, available at http://www.eu-star.at/ frameset.htm). Macroinvertebrate data were largely recorded at species and genus level, while some Diptera larvae were resolved to family/subfamily level and some taxa such as Hydracarina and Oligochaeta were recorded as such.
Our research utilised a macroinvertebrate dataset covering a 12-year period (2009-2020 inclusive) encompassing the current regions where Dh has spread in England (details in Table S1 and Figure 1) trying to minimize climate, lithological and hydromorphological background variability. To achieve this the following steps were followed: (a) the most recent distribution information for Dh in England was used (up to 2020); for sites supporting Dh, the date of the first record was used as the date of the first occurrence; (b) community data up to 3 years prior the first record was extracted and used as "pre-invaded samples"; (c) we retained "control samples" T A B L E 1 List of all the variables and descriptors tested in the study divided in taxonomic metrics, functional descriptors and abiotic characteristics

| Taxonomic, phylogenetic, and functional descriptors
A wide range of taxonomic (13), phylogenetic (2) and functional (5) descriptors were calculated and compared between pre-invaded and control samples (full details and references in Table 1). The indices comprised those routinely used by the Environment Agency for general (WHPT_TOTAL, N_TAXA-WHPT, WHPT_ASPT) and stressorspecific ecological assessments (PSI and LIFE for sediment and flow evaluation, respectively). Similar or derivate indices are widely used in river biomonitoring programs internationally (Buss et al., 2015). The communities were further characterised using the abundance and richness of EPT (Ephemeroptera, Plecoptera, Trichoptera) at family level (proxy of ecosystem conditions), abundance of Odonata, Coleoptera, Hemiptera (proxy of lentic habitats) and abundance of Oligochaeta and Hirudinea (proxy of pollution tolerant taxa).
The level of biocontamination (presence of invasive species) was explored using both the Richness Contamination Index (coded RCI) and Abundance Contamination Index (coded ACI) (Arbačiauskas et al., 2008) and the richness and abundance of other Ponto Caspian species (originating from the same geographical region as Dh, Tested in this study -Note: Codes, definitions, references and specific biological invasion hypothesis (when pertinent) are also provided. Abiotic descriptors labelled with † were measured in the field.
with specific feeding habits in the recipient community influenced the future invasibility by Dh. A taxon was considered as a predator when it was coded at least 0.5 out of 1 for its feeding habit "predator" or "piecer" within the biological traits' dataset proposed by Schmidt-Kloiber and Hering (2015).

| Abiotic descriptors
Riverine abiotic conditions were assessed by measuring four physical descriptors at each sample site: percentage of fine sediment (the proportion of clay, silt, sand), percentage of coarse sediment, the average channel width and average water depth. These latter data were available for 615 and 608 samples respectively. In addition, a buffer of 1 km-radius centred on each studied site was used to describe the local land use surface using four classes (artificial-including urban, agriculture, forested and semi-natural, wetlands and water bodies) based on the CORINE land cover dataset (using 2012 as the census year, source: https://land.copernicus.eu/pan-european/corine-landcover). A similar approach was applied by Laini et al. (2018) and Monteagudo et al. (2012) to detect the influence of land use on macroinvertebrate community and quality status of rivers. Land use analysis was performed at the site level (n = 173) within the programme QGIS (2021); this assumed that land use remained constant throughout the study period.

| Biological invasion hypotheses tested
The comparison between pre-invaded samples and control (never allowed these theories to be tested using riverine invertebrate communities for the first time in the majority of instances. The "human activity" hypothesis (also named "disturbance" hypothesis) argues F I G U R E 1 Map of study sites employed in the present research (England, UK). Control and pre-invaded are labelled in different colour [Color figure can be viewed at wileyonlinelibrary.com] that anthropogenic activities facilitate the establishment of non-native species by disturbing natural environments (Jeschke & Heger, 2018;Leprieur et al., 2008). Similarly, measuring the taxonomic and functional richness values of both control and pre-invaded sites allows the "biotic resistance" hypothesis (also called diversity-invasibility theory) to be tested. This predicts that species-rich communities should prevent or limit the establishment of new non-native species (e.g., Kennedy et al., 2002). Analysing the co-occurrence of other invaders (e.g., through considering the biocontamination signal and the presence of co-occurrent Ponto-Caspian species) allowed the examination of the "invasional meltdown" hypothesis which predicts that if multiple new species invade the same site/habitat, they may help facilitate each other's establishment (Simberloff & Von Holle, 1999). Finally, focussing on the presence and abundance of closely related species in the recipient community (e.g., number of crustacean, organisms from Gammaridae family) provided the opportunity to examine "Darwin's naturalization conundrum" that proposes that taxa with high levels of phylogenetic relatedness with invaders would reduce their chances of successful invasions (Thuiller et al., 2010).

| Statistical analysis
To determine the effect of the sample condition (control or preinvaded) on the pool of biological metrics and descriptors, a mixed model framework was used to analyse the data. Biomonitoring and functional metrics, as well as average channel width and water depth were explored using linear mixed models (LMM; package "lmerTest" Kuznetsova, Brockhoff, & Christensen, 2017). LMMs are powerful and robust tools for analysing complex datasets with multiple or clustered observations (Schielzeth et al., 2020) and were performed considering the sample condition as a fixed factor and sampling site (repeated along the temporal period) as a random factor.
Generalized linear mixed models (GLMM, glmer function) were used for variables derived from count data (e.g., abundance of crustacean, gammarids, Ponto Caspian taxa and the abundance-based metrics: EPT, OCH, OH) with a negative binomial distribution to control for over-dispersion. The percentage of riverbed substrate type, being expressed as fine or coarse (with total value equal to 100%) was examined using GLMM using a binomial distribution. Models were validated by checking the graphical distribution of residuals as well as for overdispersion (Zuur, Ieno, Walker, Saveliev, & Smith, 2009) using the "DHARMa" (Hartig, 2021) and "predictmeans" (Luo, Ganesh, & Koolaard, 2021) packages.
To investigate differences in land use typologies, models with and without spatial correlation were assessed and the final model selection performed using Akaike's information criterion. Spatial correlation was then modelled using generalized least squares and a Gaussian correlation structure. All statistical analyses were performed using R

| RESULTS
More than 650,000 organisms were identified in the dataset used in this research. A summary of all variables analysed is presented in Table 2 for pre-invaded and control groups (including mean, SD, median and maximum values).
Results from the mixed model analysis identified that many variables did not display any statistical differences between pre-invaded and control samples (e.g., LIFE, PSI, WHPT_ASPT and EPT values).
Four biomonitoring and functional metrics displayed similar values but were significantly higher in pre-invaded samples compared to control samples-WHPT_TOTAL, N_TAXA-WHPT, FRich and Fred (Figure 2, Table 3). RCI values were significantly higher for pre-invaded samples (Table 3 and Figure 3) illustrating the importance of the contemporary presence of other invaders. Similarly, the number and abundance of other alien Ponto-Caspian taxa were significantly higher at sites that were subsequently invaded than control sites (Tables 2-4). This pattern was confirmed for almost all river catchments considered ( Figure 4). In contrast, control sites (not invaded by Dh) appeared to support higher abundances of crustaceans (proxy of phylogenetic related taxa with Dh, see Tables 2 and 4).
Land uses did not differ among the 2 groups (artificial surfaces tvalue 1.43, p-value = .15; agriculture surfaces t value = 0.73, pvalue = .46, results from other typologies not shown: no value in up to 140 sites). Finally, sites that were subsequently invaded by Dh were significantly wider and deeper compared to control sites, although there were no differences associated with fine-or coarsegrained riverbed sediments (Table 5). Overall, there were few strong associations with classic biological invasion hypotheses, although supporting evidence was observed for the "invasional meltdown" and "Darwin's naturalization conundrum" (see Section 4.2 below).

| Lessons from the past: Preconditions of invasibility
The analysis of multiple community and environmental features has been recommended to better understand the observed changes associated with biological invasion (e.g., Alahuhta et al., 2019). In the current research we integrated taxonomic, phylogenetic, and functional measures of ecosystem integrity and biodiversity, as well as physical characteristics of the sites in a multidimensional approach. When these variables were analysed over the long-term study period it facilitated the comparison between the two groups of samples (preinvaded and control) located within the same geographical river catchments, but subject to different trajectories of changes (the colonisation and invasion by Dh or-control sites).
Samples from sites that were subsequently invaded by D. haemobaphes were characterized by the presence of a higher abundance of other invaders compared to control sites (e.g., Richness Con- geographical and evolutionary range provided a better indication of conditions that would favour the colonisation and invasion of Dh. Research undertaken using post invasion data, for the same areas of England, has demonstrated that different facets of biocontamination may affect a range of metrics and that an increase in the relative richness of invasive species (RCI) had a negative effect on the trajectories of multiple biomonitoring measures (Guareschi, Laini, England, Barrett, & Wood, 2021). The utility of RCI makes it a potentially important tool that scientists and natural resource managers can use to obtain a better understanding and assessment of biological invasion events, although a finer taxonomic resolution of faunal data should be used where possible (e.g., genus rather than family). Future applications in research and testing of ecological theories should explicitly incorporate both alien taxa abundance (ACI) and richness (RCI) as proxies for invasive taxa effects.

| Exploring biological invasion hypotheses applied to Dh
The dataset and analyses performed in this research allowed the consideration of several biological invasion hypotheses which have not been commonly tested using riverine invertebrates.
The "human activity" hypothesis would predict lower biomonitoring metric values and more intensive anthropogenic land-use  at sites that will experience subsequent invasion. However, our results did not support these predictions, with no differences when considering land use and higher biomonitoring metric values at sites prior to invasion. Similarly, both taxonomic and functional richness displayed higher values at sites that were subsequently invaded (compared to never invaded sites) raising doubts regarding the "biotic resistance" hypothesis and in agreement with the findings of Leprieur et al. (2008) and Henriksson, Yu, Wardle, and Englund (2015) who considered freshwater fish communities in the river and lacustrine environments, respectively. Evidence supporting the "biotic resistance" hypothesis is  (Jeschke & Heger, 2018). A global literature review found that support for major biological invasion hypotheses is uneven across taxonomic groups and habitats, although the "invasional meltdown" hypothesis was better supported by empirical evidence than other hypothesis including "biotic resistance" (Jeschke et al., 2012).
It may be possible that the successful spread of Dh is also related to other features, pathways and behaviours, not considered or quantified in the present study (e.g., presence of aquatic vegetation and specific waterway uses). Indeed, indirect measures of propagule pressure such as socio-economic activities (e.g., international cargo trade and transport) and societal awareness and lifestyles (e.g., recreational boating and fishing) as well as specific inter-basin transfers may have accidentally enhanced the rapid regional dispersal of Dh.

| Final remarks and opportunities
The detailed examination of sites prior to the invasion of Dh allowed the identification of key biotic and abiotic characteristics that may help in identifying sites at future risk of invasion by this highly successful amphipod. Sites that would be invaded by Dh in the future were characterized by higher abundances of other pre-exiting invaders, lower abundances of other crustaceans and the sites typically had a greater channel width and water depth. Sites sharing these features could be monitored in the future, in England and elsewhere where Dh has been widely recorded (e.g., Labat et al., 2011), to accurately monitor and if possible prevent its future range extension. Recognizing specific abiotic and biotic characteristics that may precondition a site/community for subsequent successful invasion is essential for understanding and better predicting the future trajectories of community change.
Overall, our findings, using Dh as model organism, provided evidence to support the "invasional meltdown" and, at least partially, "Darwin's naturalization conundrum" based on the variables analysed.
However, biological invasions within riverine ecosystems are still less studied compared to other ecosystems and the challenges of working in water (where most organism cannot be readily observed) make them more difficult to detect and to develop generalizable concepts.
Traditional ecological theories are a crucial starting point for data analysis, and riverine systems will benefit from further hypothesis assessment (both experimental and observational) in the field of biological invasion (e.g., multiple co-occurrent invaders) as well as new contextspecific approaches specifically designed to consider freshwater ecosystems and the communities they support.