Behavioural responses to environmental novelty in demersal, cover-seeking invasive fish and their native counterparts

Description

Invasive species may differ from native species in terms of behavioural responses to the stress of encountering a novel environment. Learning about the nature of these differences can help to understand the mechanisms of dispersal and success of the alien species in colonised environments. Here, we have investigated this topic using the Ponto-Caspian gobies as model species. We compared the behaviour of two invasive goby fishes (the racer goby Babka gymnotrachelus and the monkey goby Neogobius fluviatilis) to that of their native counterparts (the European bullhead Cottus gobio and the gudgeon Gobio gobio, respectively). We used three laboratory tests to measure boldness-shyness traits: shelter occupancy test, novel object test and open field test. The fish behaviours were analysed using Noldus Ethovision XT ® 10.1 program. The uploaded files contain raw data exported from the program. There is a “description” sheet in each file to explain the variable meanings. We performed a Principal Component Analysis (PCA) on the correlation matrix separately for each test and each pair of fish species (six analyses in total) using these data. Then we run general Linear Models on the principal components extracted by the PCA to get the results. The results showed that European bullhead left the shelter later and was less active, as well as avoided the open field to a greater extent than the racer goby. The gudgeon was more associated with the shelter and novel object than the monkey goby and, in contrast to the monkey goby, decreased its activity in the presence of the novel object and in the open field. All the species were attracted to the vicinity of the novel object. Our study suggests that the invasive Ponto-Caspian gobies are bolder when confronted with structural changes in their environment, and have a greater potential to spread across the open bottom, devoid of hiding places, compared to their native analogue species.

Steps to reproduce

We performed a Principal Component Analysis (PCA) on the correlation matrix separately for each test and each pair of fish species (six analyses in total) using behavioural variables from Table 1 (except for latency variables in the open field test, see below). The PCA allows to reduce problems with multiple statistical comparisons and avoid the multicollinearity of multiple independent variables by producing orthogonal components based on sets of intercorrelated raw variables. Principal components (PC) were extracted based on their eigenvalues greater than 1. We took into account the original variables with absolute values of their loadings greater than 0.5 after Kaiser-Varimax rotation when explaining the meanings of the particular PCs. The normality and homoscedasticity assumptions of determined PCs were not violated based on visual inspection of residual plots, thus we decided to use parametric tests for their further analysis. For the shelter occupancy test, we used a 2-way General Linear Model (GLM), with Species (one of the two species in the pair) and Light conditions (light or darkness) as between-subject factors. Principal components from the novel object test were analysed using a 3-way GLM with Species, Light conditions and Treatment (control/object) as between-subject effects. For this test, we only interpreted terms involving the novel object presence effect as indicating responses of fish to the introduction of the novel object. For the open field test, we used a 3-way General Linear Mixed Model (GLMM), where Species and Light conditions were between-subject fixed effects, Period (early or late response) was a within-subject fixed effect and an individual ID was a random effect. For the open field test, we analysed the latency variables (time to the first occurrence of an event) separately, using the Cox Proportional Hazards model, as censored observations were present in the dataset. Moreover, these variables were valid only for the early response period. For the novel object and shelter occupancy tests, we included latency variables to the PCA. For each analysis, we started with the full factorial model, then we simplified it by dropping consecutively the highest-order non-significant interaction terms. Based on the Akaike information criterion, we retained a more complex model when its AIC was lower than that of the simpler model by 2 or more (Burham and Anderson 2002). All models were then followed by LSD post-hoc tests with Bonferroni corrected p-values for significant terms. All the analyses were run in R 4.1.1 (R Core Team 2021). The Cox Proportional Hazards model was run using the “survival” package (Therneau 2022), GLMM’s using “lmerTest” package (Kuznetsova et al. 2017) and post-hoc tests using “emmeans” package (Lenth 2022).

Institutions

Categories

Funding

Licence