Selection on fish personality differs between a no‐take marine reserve and fished areas

Abstract Marine reserves can protect fish populations by increasing abundance and body size, but less is known about the effect of protection on fish behaviour. We looked for individual consistency in movement behaviours of sea trout in the marine habitat using acoustic telemetry to investigate whether they represent personality traits and if so, do they affect survival in relation to protection offered by a marine reserve. Home range size had a repeatability of 0.21, suggesting that it represents a personality trait, while mean swimming depth, activity and diurnal vertical migration were not repeatable movement behaviours. The effect of home range size on survival differed depending on the proportion of time fish spent in the reserve, where individuals spending more time in the reserve experienced a decrease in survival with larger home ranges while individuals spending little time in the reserve experienced an increase in survival with larger home ranges. We suggest that the diversity of fish home range sizes could be preserved by establishing networks of marine reserves encompassing different habitat types, ensuring both a heterogeneity in environmental conditions and fishing pressure.


| INTRODUC TI ON
Fishing-induced evolution and the consequences for populations have now been extensively documented (Kuparinen & Festa-Bianchet, 2017). For example, selective fisheries may alter life-history traits in a population by causing a shift towards maturation at earlier ages and smaller body sizes (Kuparinen et al., 2016;Olsen et al., 2004).
However, fishing-induced evolution of behaviour has received far less attention (Diaz Pauli & Sih, 2017). Interestingly, growth rate can be related to behavioural expression, and a selection regime targeting larger individuals may reduce the overall boldness in the population compared with a selection regime where small individuals are targeted (Biro & Post, 2008;Uusi-Heikkilä et al., 2015). Harvesting may also select directly on behaviour (Uusi-Heikkilä et al., 2008). For example, passive fishing gear can select against traits such as strong diel vertical migration (Olsen et al., 2012) and large home ranges (Alós et al., 2016) and lead to increased timidity , while active fishing gear such as trawling may favour bolder individuals (Andersen et al., 2018;Diaz Pauli et al., 2015). Moreover, since the vulnerability to certain harvest conditions may vary from one fish 2 | MATERIAL S AND ME THODS

| Study species
The brown trout (Salmo trutta) is a salmonid species in which anadromous populations are called sea trout. Sea trout have a highly variable life history with some individuals spending only the summer at sea, while others spend most of their time in marine areas only returning to the river to spawn during fall (Klemetsen et al., 2003). Marine migrations are motivated by access to more food, with important trade-offs being adjustment to different salinities, increased energetic cost of movement and a potentially higher predation risk (Thorstad et al., 2016). The balance of these trade-offs is likely an important part of the explanation for the diversity of migration strategies within populations (Thorstad et al., 2016) and population differentiation between streams (Knutsen et al., 2001;Olsen et al., 2006). In Norway, fishing for sea trout in marine habitats is recreational and permitted all year. Fishing can only be done using hook-and-line, except for one month in summer where specialized traps are permitted in the southern part of Norway. The minimum legal size for sea trout in the marine habitat in Norway is 35 cm. In the fjord, potential predators of sea trout are, among others, gulls, cormorants (Phalacrocorax carbo), harbour seals (Phoca vitulina) and gadids, as reported from a study system in western Norway (Jonsson & Jonsson, 2009).

| Study system and data collection
Movement data were collected in the Tvedestrand fjord (3.8 km 2 , max depth: 87 m) located in southern Norway between spring 2013 and fall 2017 (Figure 1). A telemetry array consisting of 50 Vemco VR2-W receivers (VEMCO Ltd.) was deployed in the fjord, with the receivers being attached to moorings and kept at three metres depth aided by subsurface buoys (for more details, see Villegas-Ríos, Réale et al., 2017).
One receiver was located close to the spawning river, Østeråbekken, in order to monitor river migrations, and three receivers located in the outer part of the fjord served as a gate to detect individuals dispersing towards the outer fjord and sea areas. The high density of receivers ensured a good coverage of the fjord (see also Supporting Information, Thorbjørnsen et al., 2019). A marine protected area (1.5 km 2 ) prohibiting all types of fishing was established within the spatial coverage of the telemetry array in 2012. Fishing is also prohibited in Østeråbekken and up to 100 m from the outlet of the stream.
Sea trout were caught around the centre islands of the fjord in 2013 (April: n = 3; May: n = 26; September: n = 24; November: n = 7), 2015 (June: n = 3; October: n = 14; November: n = 5) and 2016 (April: n = 4; May: n = 7) using a beach seine, and also by electrofishing in the spawning river at 11 November 2016 (n = 23). Beach seine was chosen in an attempt to minimize sampling-induced selection of particular behavioural types (Olsen et al., 2012). Electrofishing was added as a complement to increase sample size in 2016. Individuals were anaesthetized with clove oil, and a transmitter was inserted in the abdominal cavity (for details, see Olsen et al., 2012). We used Vemco V9P and V13P transmitters, which had a maximum battery life of 508-696 and 1292 days, respectively. Signals were emitted with a random delay of 180 ± 70 s. Accuracy and resolution of depth measurements were ±2.5 m and 0.22 m, respectively, and max depth was 50 m or more for the different tags. Sea trout were not externally marked. Fin clips were taken for DNA analysis and preserved in 95% ethanol.
In total, 116 sea trout (mean body length: 337 mm, range: 215-635 mm) were caught, tagged and monitored in the Tvedestrand fjord during a 1669-day study period (spring 2013-fall 2017). A total of 20 individuals were excluded from the study due to tag malfunction (n = 4), post-surgical mortality (n = 5) or limited presence in the study area (<14 days, n = 11). Time spent in the study ranged from 1 to 20 months. Initial data exploration revealed that sex had no effect on any behavioural trait.

| Data preparation and estimation of behavioural metrics
Detections were downloaded from the receivers and processed using the VUE software (VEMCO Ltd.), and further data preparation and analyses were done in the R environment (R Core Team, 2016). All detections after presumed death were censored, which was defined to have occurred when continuing detections indicated that horizontal and vertical movement had ceased (Olsen et al., 2012). Note that this could also represent transmitter loss. Fish were defined as dispersed after having followed a directional path out of the reserve with final detections occurring at the outermost receivers. Single detections within one day per fish were removed to eliminate potential code collisions and false detections, and above surface depth measurements were defined as NA. Four traits were used to describe the movement behaviour in the marine phase: home range, mean swimming depth, activity and diurnal vertical migration. Monthly replicates were used for all traits. Monthly 95% home ranges were calculated using locations based on position averages (PAVs, centres of activity), following Simpfendorfer et al. (2002). PAVs are weighted average locations within an array of receivers, based on the number of detections at each receiver during a specified time period (Simpfendorfer et al., 2002), in this case 30 min. Home ranges were then calculated from PAVs using kernel utilization distributions (bandwidth = 60, extent = 0.5) using the adehabitatHR package in R (Calenge, 2006). Depth measurements were averaged over months after removing replicated measurements occurring when a signal is detected at more than one receiver.
Following Freitas et al. (2016), activity was defined as short-term changes in depth and this was approximated as the standard deviation of depth per hour and then averaged over months. Diurnal vertical migration was calculated as the difference in mean depth from day to night within a calendar day and then averaged over months. Day and night phases were defined by solar elevation data obtained from the National Oceanic & Atmospheric Administration (NOAA) through the maptools package in R (Bivand & Lewin-Koh, 2018). Only months where the fish was present in the fjord for a minimum of 15 days (not necessarily consecutive) were included in analyses.

| Repeatability estimation
Univariate mixed-effects models were fitted for each behavioural trait using the nlme package (Pinheiro et al., 2018) in R. For modelling purposes, home range and activity were log-transformed to meet normality assumptions of the residuals. Monthly averages of each behavioural metric served as replicates for individual fish and individual sea trout identity was included as a random effect. We considered a trait to be repeatable when the inclusion of the random effect significantly improved the model fit. Provided that the random effect was supported, repeatability was calculated following Dingemanse and Dochtermann (2013) as: where V ind 0 is the among-individual variance and V e 0 is the withinindividual variance. Model selection was done in two steps: (1) selecting the overall model structure by assessing if including Map of the Tvedestrand fjord (below) and its location along the Norwegian Skagerrak coast (above). The marine reserve in the centre of the fjord is delineated with black lines. Blue dots represent receiver locations, and red dots represent capture locations the identity of the fish as a random effect and temporal autocorrelation between months improved the model (method = restricted maximum likelihood), followed by (2)

| Survival
A survival curve was generated by computing a Kaplan-Meier estimator for right-censored data (Cox & Oakes, 1984) using the 'survival' package in R (Therneau, 2015).

| RE SULTS
Home range size was a repeatable movement trait (repeatability = 0.21, Table S1), while mean swimming depth, activity and diurnal vertical migration were not (Tables S2-S4). Mean monthly home range size was 0.407 km 2 (range: 0.065-2.14 km 2 ), increased with body length, and was larger for fish caught in the fjord than fish caught in the river (Table 1, Figure S1). Home range size was also affected by season, being the largest in spring, followed by fall and summer, and the smallest in winter (  Figure S1). Mean swimming depth was also affected by an interaction between fish body length and season. Mean swimming depth increased with body length and differed between seasons, with fish being located at more shallow depths during fall compared with all other seasons. The interaction between body length and season indicated a stronger positive effect of body length on mean swimming depth in summer, followed by spring, winter and fall (Table 1).
Activity (mean = 0.47 m, range: 0.018-3.67) increased with body length and was higher for fish caught in the fjord (Table 1, Figure S1).
Activity differed between the seasons, and fish were most active during spring and summer, and least active during fall and winter (Table 1).
Diurnal vertical migration, the difference in mean depth from day to night (mean = 0.95 m, range: −0.75 to 5.08), was larger for fish caught in the fjord than fish caught in the river ( Figure S1) and was affected by an interaction between body length and season (Table 1). Diurnal vertical migration increased with body length and differed between seasons, with fish having a larger daily movement span during spring and summer than in winter and fall. The interaction between body length and season indicated a stronger positive effect of body length on diurnal vertical migration in spring and summer than in winter and fall (Table 1).
Including autocorrelation led to significant improvement of all models with a behavioural trait as the response variable (Tables   S1-S4).

| D ISCUSS I ON
Sea trout revealed individual consistency in home range size over a period of several months or even years, reflecting that home range can be considered an aspect of personality. Further, we found that home range size affecwted survival, and this relationship differed depending on the proportion of time the fish spent inside the reserve.
For individuals that spent more than 48% of their time in the reserve, larger home ranges were associated with decreased survival, while individuals that spent less than 48% of their time in the reserve TA B L E 1 Summary of selected linear mixed-effects and lme models explaining movement behaviour in sea trout experienced increased survival with increasing home range size. In other words, the fitness landscape of sea trout appears to be influenced by spatial management, here represented by a no-take marine reserve. As discussed below, this suggests that fish behaviour might evolve in response to fishing and therefore certain fishery management measures.
We found that home range size had a repeatability of 0.21, indicating that 21% of the variation in home range size is variation that occurs among individuals. This is comparable to the behavioural trait mean repeatability of 0.37 overall and 0.32 for fish previously reported in a meta-analysis by Bell et al. (2009)  For sea trout that spent less than 48% of their time in the reserve, survival increased with increasing home range. Note that this is the opposite pattern as previously found by both Alós et al. (2016) who reported selection against large home ranges in harvested areas, and Härkönen et al. (2014) finding that being explorative is linked to an increased vulnerability to angling in brown trout. Overall, there is contrasting evidence on the relationship between home range and surviving the fishery. Monk and Arlinghaus (2017) report no effect of swimming distance or activity space on vulnerability to capture by angling, while Olsen et al. (2012) report increased fishery survival for fish that displayed extensive horizontal shifts. Interestingly, individuals that express more risky behaviours in the wild, including exploration, experience increased survival (Moiron et al., 2020). Risky behaviours may lead to acquiring more resources followed by an increase in natural survival (Moiron et al., 2020). Having a larger home range size may be one such risky behaviour. Furthermore, it is less obvious why a large home range mainly located outside the reserve yields better survival than a large home range mainly located inside the reserve. A study with replicated protected-unprotected area pairs could help to investigate whether the patterns found in this study are general patterns. These findings, combined with the fact that home range is repeatable and therefore likely partly heritable (Dochtermann et al., 2015), may entail evolutionary consequences for populations that are partially protected by marine reserves. That said, any local evolutionary change will also depend on the level of gene flow (Lenormand, 2002). In accordance with estimates of how much additive genetic variation contributes to personality, Dochtermann et al. (2015) estimated that the ratio of heritability to repeatability collected from literature averaged at 0.52 and ranged from 0 to 0.96 (Dochtermann et al., 2015). Few studies investigate heritability of behaviour in sea trout, but in a laboratory study on adfluvial brown trout, Kortet et al. (2014) found heritability of 0.14 (± 0.096) for the stress response 'tendency to freeze', but no heritability for boldness, exploration and aggression. To the best of our knowledge, there are no estimates of heritability of sea trout behaviour in the wild. Our paper is the first to present estimates of repeatability of sea trout behaviour in the wild, indicative of additive genetic variation (Dochtermann et al., 2015). In addition to additive genetic variation, repeatability may also reflect learning (Adriaenssens & Johnsson, 2011) and individual variation in utilizing heterogeneous environments (Bell et al., 2009). Hence, repeatability of home range size may also, to some degree, reflect individual differences in habitat use (Bell et al., 2009).
Body length affected all movement traits, with larger fish having larger home ranges, utilizing a larger range of depths and having a higher activity. As survival was affected by home range size, this may imply correlated selection on body length. However, body length did not affect survival directly. Home ranges were the largest in spring, and fish were more active during spring and summer than fall and winter. This is in accordance with sea trout intensifying their food search as temperatures increase during spring and summer (Klemetsen et al., 2003). Fish also swam deeper during spring and summer, which can be associated both with different habitat use and that the trout seek out colder water temperatures optimal for growth when surface temperatures rise (Eldøy et al., 2017;Kristensen et al., 2018). Fish tagged in the sea had larger home ranges, utilized a larger range of depths and had higher activity than fish tagged in the river.
Median survival in the wild was close to 11 months, and survival was higher for fish tagged in the fall. The latter could be explained by the upcoming spawning ascent, where sea trout will receive protection from fishing and experience a lower predation risk in the river (Thorstad et al., 2016). In general, trout survival is higher in freshwater as compared to sea (Solomon, 2006), and the duration of migration varies both within populations and among populations and latitudes (Klemetsen et al., 2003). This implies that yearly survival will vary substantially between river systems.
Return rates from 193 sea trout tagged in the nearby river Storelva (<5 km from our study system) revealed 40% survival for trout spending one or two years at sea (Haraldstad, 2015). Survival might have been underestimated due to tag excretion, which would have led individuals to be falsely defined as dead. Also, there might be a negative effect of tagging on survival. A study on gastrically tagged salmonids found that small (9 mm) and large