Increased exposure to pathogens is often considered to be one of the most significant costs of group living. As a result, animals typically avoid close association with individuals that manifest symptoms of disease. The question remains, however, whether avoidance behaviours are mediated by effects relating specifically to the disease itself, or through recognition of more general sickness behaviours that emerge from the activity of the immune system in response to infection. Here we examined the effects of an immune challenge, induced through exposure to bacterial lipopolysaccharides (LPS), on individual and social behaviour in female guppies (Poecilia reticulata). We found that in a binary choice presentation, healthy focal individuals showed a strong preference to associate with stimulus shoals of saline-injected fish rather than stimulus shoals of LPS-injected fish.  In a subsequent examination of group-level dynamics, shoals of LPS-injected females were more dispersed than similar shoals comprised of control fish, and showed less coherent collective behaviour despite the lack of any obvious difference in general activity between LPS and control fish. We discuss these findings in relation to the need for social animals to mitigate the risk of disease that is associated with living in proximity to conspecifics.

Study Species

We used the guppy (Poecilia reticulata) as our study species because it is a widely used model species in behavioural ecology and physiology studies. For the present study we used only females because of their greater shoaling tendency and to avoid the confounding effects of courtship and mating behaviour. Fish used in the present study were adult females measuring 20-30 mm sourced from a feral population of guppies in Darwin, NT, Australia. Fish used in the experiments were housed in 90L aquaria where they were kept at 25°C on a 12:12 light:dark cycle. Fish were fed daily ad libitum with commercially available fish flakes (Nutrafin). 

Immune Challenge

Females were randomly assigned to either LPS or control groups (LPS-injected: N = 60, Control: N = 60). All fish were lightly anaesthetised using a clove oil solution (50mg/L) and then weighed at the time of injection. Fish assigned to the LPS treatment were injected intramuscularly in the dorsal region at a dosage of 10mg LPS (Sigma-Aldrich; Lipopolysaccharides from Escherichia coli serotype O111:B4) per gram of fish. LPS was dissolved in a solution of Milli-Q^® water and salt to a standard saline concentration of 0.9%. Control fish underwent the same procedure but were injected with saline only. Following injection, fish were immediately transferred to individual tanks containing aged tap water where they recovered within 1-2 minutes. Following injection, fish were monitored for signs of stress for 48 hours prior to experiments. A 48 hour interval between injection and testing was chosen on the basis of precedent from previous published studies which show that the immune response peaks 48hrs post-injection in a closely-related species (Gambusia holbrooki). 

Experiment 1: Binary Choice

Trials were conducted in a white rectangular Perspex arena measuring 60x30x20cm, which was filled to a depth of 10cm. The arena was divided into three zones, comprising a central chamber which measured 34x30cm as well as two chambers at opposite ends of the longest axis of the arena each measuring 13x30cm. Focal individuals were introduced into the central chamber, while stimulus shoals were placed in each of the end chambers. The barriers separating stimulus fish from the focal individual were constructed using transparent perforated plastic (PennPlax) to facilitate the transmission of visual and chemical cues. Each stimulus shoal, comprised of 6 fish, were injected intramuscularly with either saline or LPS 48 hours prior to testing. Focal fish were healthy individuals that had not exposed to either the LPS or control (saline) treatment prior to testing. Stimulus shoals were introduced to their designated chamber of the arena and allowed to acclimate for 1 hour. After this, a single focal fish was introduced to the central chamber of the arena and filmed for 10 minutes. The focal fish was then removed, and after 5 minutes a new focal fish was added and the procedure repeated. We conducted a total of 4 trials before replacing stimulus shoals. In total 8 sets of stimulus shoals were used for 4 trials each for a total of N=32 trials. There was one focal fish per trial (N=32), none of which were reused. Only the last 7 minutes were extracted for analysis in order to allow the focal fish an acclimation period of 3 minutes. The extracted video footage was then reviewed and the time each focal fish spent in association with each of the stimulus shoals was manually recorded. Association of the focal fish with a stimulus shoal was defined as the time (in seconds) spent within 5cm of the PennPlax barrier separating the central chamber from each of the end chambers containing the stimulus shoals. Focal fish which did not investigate both stimulus shoals during the trial were excluded. Only N=1 trial was excluded from analysis on this basis (the focal fish associated solely with the LPS-injected shoal). For completeness, a further analysis which included the outlying individual was conducted, however the statistical outcome and interpretation of the test was not affected. (For comparison, the statistical outcome without the outlier was χ²= 29.048, Df = 1, P < 0.001, while with the outlier included, the outcome was χ²= 18.375, Df = 1, P < 0.001.)

Data Analysis

Statistical analysis was performed using R (v4.1.2; R Core Team 2022). Data were examined using Q-Q plots and Shapiro-Wilk tests, and satisfied the assumption of normality. We analysed the time in seconds spent by focal fish with each of the two stimulus shoals using a linear mixed effects regression model, using the package lme4 (Bates et al., 2015), specifying the treatment of the stimulus shoals as the fixed effect. Focal fish ID and the stimulus shoal batch (from 1-8) were included as random factors in the model.

Experiment 2: Collective Behaviour

At 48hrs post-injection fish were transferred from their individual holding tanks to a circular arena with a diameter of 70 cm and a depth of 5 cm. Arenas were lit using cool white LED strips (6500K) and surrounded with white screens to minimise external disturbance. All trials were filmed using a Canon G1X camera positioned 1.1m above the arena at a frame rate of 24fps and a resolution of 1920p. Each trial was filmed for 12 minutes from the time of introduction. Following this, fish were removed and transferred to new holding tanks. No fish were reused. 

We conducted two treatments, each involving test groups of 6 fish. In the wild, female guppies are social and tend to live in small shoals of between 2 and 10 individuals.  As such, a group size of 6 was selected on the basis of being ecologically realistic. The treatments were LPS, in which all fish were LPS-injected and Control, in which all fish were saline-injected. We performed (N=10) per treatment to a total of 20 trials.

Data Extraction and Analysis

For each group trial, we extracted the last 10 minutes of video footage for analysis. This allowed the fish an initial 2-minute acclimation period prior to data collection. Videos were tracked using TRex; x,y coordinates for each fish over the 10-minute test period were extracted and we used these values as the basis for subsequent analyses. 

To examine the effect of LPS on individual activity and group cohesion, we calculated mean individual speed and mean nearest neighbour distance for each fish. Again, data were analysed using R (v4.1.2; R Core Team 2022). Data were examined using Q-Q plots and Shapiro-Wilk tests of normality. These dependent variables were analysed using linear mixed-effect models package lme4 with treatment as the independent variable and specifying group as the random factor.

To examine the effects of LPS on the distribution of near neighbours relative to a focal fish, and to characterise the local interactions between that focal fish and its neighbours, we used the methods described in detail in Schaerf et al., (2016) and which we previously employed in Encel et al., (2021) Specifically, we examined the local distribution of near neighbours, and the alignment, speed, change in speed and change in heading of a focal individual relative to the positions of near neighbours. To achieve this, we first transformed the position of each fish to a consistent coordinate system where a focal individual was located at the origin (0,0), and the direction of motion of the focal individual was parallel to the positive x-axis. A square domain centred on the focal individual was then sub-divided into smaller overlapping square bin regions. The larger square domain extended over the region where − 100 < x ≤ 100, − 100 < y ≤ 100 millimetres, that is, out to distances of approximately 4 body lengths to the front and back, and to the left and right, of the focal individual. The smaller square bin regions had side lengths of 15 mm (approximately half a body length for the guppies), with the leftmost and/or bottommost edges of adjacent bins separated by 3.75 mm (one-quarter of a bin width). Square bins and the rectangular (x, y) coordinate system were chosen so that each bin covered equal areas of the domain near the focal individual (as opposed to bins based on distances to, and angular ranges containing groupmates, where the area of bins would grow as the distance from the focal individual increased). The overlap of the bins is a means to smooth the resulting plot, analogous to the moving window used to calculate a moving average. The number of times that neighbours occupied each bin was counted, with data aggregated by treating all group members as the focal individual in turn. Absolute counts in each bin were then normalised by dividing by the total counts across all bins.

We adapted the randomisation method based on the mean absolute difference between pairs of fitted functions (Schaerf et al., 2021) for across group comparisons to examine if there was a significant difference across treatments. We adjusted the thresholds for significance for the mean absolute difference randomisation tests according to the Holm-Bonferroni method (Holm 1979) to take into account multiple pairwise comparisons.