Animals show among-individual variation in behaviours, including migration behaviours, which are often repeatable across time periods and contexts, commonly termed “personality”. These behaviours can be correlated, forming a behavioural syndrome. In this study, we assessed the repeatability and correlation of different behavioural traits i.e., boldness, exploration, and sociality and the link to feeding migration patterns in Atlantic cod juveniles. To do so, we collected repeated measurements within two short-term (three days) and two long-term (two months) intervals of these personality traits and genotypes of the Pan I locus, which is correlated to feeding migration patterns in this species. We found high repeatabilities for exploration behaviour in the short- and long-term intervals, and a trend for the relationship between exploration and the Pan I locus. Boldness and sociality were only repeatable in the second short-term interval indicating a possible development of stability over time and did not show a relation with the Pan I locus. We found no indication of behavioural syndromes among the studied traits. We were unable to identify the existence of a migration syndrome for the frontal genotype which is the reason that the link between personality and migration remains inconclusive, but we demonstrated a possible link between exploration and the Pan I genotype. This supports the need for further research that should focus on the effect of exploration tendency and other personality traits on cod movement, including the migratory (frontal) ecotype to develop management strategies based on behavioural units, rather than treating the population as a single homogeneous stock.

The methods followed the same protocol as described in Beukeboom et al. 2022; 102 age 0+ cod juveniles (weight range = 0.75-4.39 g and mean = 1.87 g; standard length range = 3.83-7.55 cm and mean = 5.72 cm) were beach seined from the 3^rd to 12^th of October 2019 in three different fjords around the Westfjords of Iceland to obtain genetical variation. They were transported to a laboratory in Bolungarvik, Iceland and housed individually in 9.5-litre tanks (~29x21x19cm, water level 16 cm, Aquaneering Inc.). The recirculating system contained freshwater mixed with marine salt to achieve a natural salinity of 30 ± 2‰, a temperature of 11 ± 1°C (November) and 10 ± 1°C (December-June), ammonia levels of <0.5 ppm, oxygen levels of 10.4 ± 0.1 mg/L and a constant photoperiod of 12 hr:12 hr (7 AM-7 PM GMT). The water circulated through the Aquaneering system, passing through all the tanks, a biofilter, sieves (mesh size 25 µm), and a UV light for sterilization. Every tank had a grey PVC pipe to provide shelter to the fish. Fish were fed twice a day alternating defrosted shrimp and bloodworms daily ad libitum. On the experimental days, feeding took place after the experiment to avoid any differences in feeding motivation. After the first trial of the open field experiments every month, the fish were measured for weight and standard length. In November, a set of trial experiments was performed which are presented in Beukeboom et al., 2022.

Behavioural tests

Of the initial 102 caught individuals, 43 were included in the analysis (see Results). Each individual underwent a cycle of a shelter test (ST, boldness), open field test (OFT, exploration), novel object test (NOT, boldness), and mirror test (MT, sociality) during four trials: January 19-21 (Trial A), and January 22-24 (Trial B), March 15-17 (Trial C) and March 18-20 (Trial D) (figure 1). All fish started with the ST, followed by the OFT. Consecutively, 20 fish were subjected to first the NOT and then the MT, while the rest (N=23) received the MT first and then the NOT to take any influence of the NOT or MT test order into account (figure 1). In addition, all fish were tested in the same overall order to standardize the intervals between the tests for each fish. Data in the ST was collected manually, while OFT, NOT, and MT videos were analyzed with video-tracking software (Ethovision, Noldus, v. 15.0). A smoothing parameter of 0.1 was set in the video-tracking software, which set the sample points to the previous location until the distance moved was more than 0.1 cm which removed any noise of moving pixels due to imperfect light conditions, as this was found to be appropriate for the setup we used (Beukeboom et al., 2022). The water temperature of the experimental tank ranged from 9.2-11.5°C.

Shelter test (ST)

The shelter test is commonly used as a measure of boldness (see for an overview Toms et al. (2010)), where bolder fish have a shorter exit time. The fish was gently captured with a dipnet out of its home tank and placed in a shelter (26L; 40x40x40 cm, water level 16 cm) and after five minutes of acclimation, the door was lifted. The fish was given five minutes to leave the shelter, and latency to exit was recorded. If it did not leave voluntarily, it was gently forced out using a dip net into the arena (51L; 80x40x40cm, water level 16 cm) and given a maximum score of 300 seconds. Unfortunately, in 95 out of 158 ST trials (60.1%), fish did not leave the shelter voluntarily. This was most likely due to the test duration being too short, which resulted in insufficient variation in this behaviour. This original boldness measurement was therefore excluded from further analysis, but a binary factor (left shelter y/n) was included in our models to account for any influence of the fish being pushed out of the shelter rather than leaving voluntarily.

Open field test (OFT)

In absence of a predator, juvenile cod have been shown to move freely across an open space (Nordeide & Svåsand, 1990), which most likely serves the function of information gathering (Hughes, 1997). Therefore, the OFT was used as a proxy for exploration, where it is expected that explorative fish swim greater distances and cover a bigger area than non-explorative individuals. The test was carried out as follows: as soon as the fish entered the arena during the ST (51L; 80x40x40cm, water level 16 cm, figure 1), the shelter was closed and the fish was video recorded for five minutes. The total distance travelled and the total area covered (i.e. traversed, calculated by using the total unique x/y coordinates rounded to the nearest integer) were extracted. Area covered and distance travelled (both log-transformed) were highly correlated (PEARSON: r = 0.97; CI 95% [0.96 to 0.98]; P < 0.001). Because the area covered is statistically ceiled (i.e., there is a maximum number of unique X/Y coordinates available), the total distance travelled was used in further analysis, to catch the maximum variation possible.

Novel object test (NOT)

The novel object test was used as a measure of boldness, where shy individuals are expected to flee, retreat, or are inactive, while bold individuals are expected to become more active and approach the novel object (Toms et al., 2010). The NOT was carried out after the OFT (N=20) or after the MT (N=23). When present, the mirror was removed, and a novel object was dropped in the middle of the arena (Figure 1). In January, this was a red, tin can (⌀ 6 cm, 113 cm²) and in March this was a blue plastic pipette tip rack (13x10 cm; 117 cm²) to reduce habituation to the object. Five minutes were video recorded and the mean distance to the object was extracted as a proxy for boldness.

Mirror test (MT)

Cod juveniles can show plastic social behaviour, from shoaling to aggression (Meager et al., 2018). The mirror test can be used to measure both sociality and aggression, depending on the species and developmental phase. While Villegas-Ríos et al. (2018) used the mirror test as a measure of aggression in adult cod, the life stage where the risk of being predated is minimal, we assume that the mirror test in this study on juveniles elicited social behaviour instead for two reasons. Firstly, the arena was open with no shelter to hide, which has been shown to elicit shoaling behaviour in juvenile cod (Laurel et al., 2004). Secondly, aggressive/submissive behaviour often occurs when opponents differ in size (McCormick & Weaver, 2012; Sverdrup et al., 2011), and as the mirror just reflects the fish itself, these size differences are non-existent. The MT was carried out after the OFT (N=20) or after the NOT (N=23). When present, the object was removed, and a mirror was placed opposite the shelter door (Figure 1). Five minutes were video recorded, and the total time spent in a 10 cm zone in front of the mirror was extracted as a proxy of sociality. After the experiment, visual inspection of the videos for aggressive and/or social behaviours confirmed our decision; social behaviours, such as repeated approaches to the mirror at cruising speed and “hanging around” the mirror were highly represented, and aggressive behaviours such as accelerations towards the mirror, biting or c-shaping were absent (Sverdrup et al., 2011).

Genotyping

In June 2020, at the end of the overall project, the fish were euthanized with phenoxyethanol (1.6mg/L) and fin clips were taken to assign the fish to either coastal (Pan I^AA), frontal (Pan I^BB), or heterozygote (Pan I^AB) using PCR analyses as described in Pampoulie et al. (2006).

Statistics

All data were analyzed using the same method as in Beukeboom et al. (2022) using R (v. 4.1.2; R Core Team, 2021). We fit multivariate linear mixed models to estimate the repeatabilities of and the correlations between the total distance travelled (OFT), mean distance to object (NOT), and time spent in mirror zone (MT). The models were fit using the Bayesian software Stan (Carpenter et al., 2017) run via the ‘brms’ package (Bürkner, 2017). We ran four separate models containing four different subsets of the data: short-term (∆ three days) for January (trials 1 and 2; SJ) and March (trials 3 and 4; SM) and long-term (∆ two months) for the first trials (trial 1 and 3; LA) and the second trials (trial 2 and 4; LB) of each month. The models simultaneously regressed each dependent variable (i.e., OFT, NOT and MT estimates) against a set of fixed and random effects while also quantifying the covariance between the dependent variables. The fixed effects of weight (g), standard length (cm), Fulton’s condition factor (K= Weight /Length³ × 100), and specific growth rate (SGR= ∆ ln(weight) *100 / ∆ day) that could influence the personality estimates were evaluated for collinearity. Pearson correlations revealed that all measurements were substantially correlated (Figure A1) and therefore only SGR was included in the analysis. The three personality estimates and SGR were scaled using z-scoring (subtracted the mean and divided by the standard deviation) separately for the subset of data used in each model. The full version of each model was fit with the scaled personality measurement as response variables (i.e. total distance travelled, mean distance to object, and total time in mirror zone), the four fixed effects of genotype, SGR (since the previous month for short-term, January-March for long-term), trial and shelter leave (y/n). We include a binary covariate to indicate the order in which fish were tested for the NOT and MT. The random-effects structure included individual fish identity (ID) as a grouping variable, allowing us to calculate the repeatability of the personality estimates as the ratio of the among-individual variance and the sum of the among-individual and residual-level variances (Johnson & Koch, 2011). Moreover, the model estimated covariances between the personality estimates at both the ID and residual levels. The among-individual covariance quantified the degree to which the personality estimates were correlated among individuals across multiple trials (i.e. behavioural syndrome), while the residual level covariance quantified the degree to which the personality estimates were correlated among observations independently of the identity of individuals. The model was run for 4000 iterations (2000 for warmup and 2000 for sampling), four chains, an adapted delta of 0.9 and all other parameters set to their defaults. Convergence was assessed using the standard diagnostics provided by Stan (Bürkner, 2017; Carpenter et al., 2017), including the potential scale reduction factor ( ), effective sample size, and visual inspection of trace plots and histograms for each model parameter. We used medians for point estimates and quantiles with 95% coverage for uncertainty intervals (UI_95%).