Individual behavioral variation reflects personality divergence in the upcoming model organism Nothobranchius furzeri

Abstract In the animal kingdom, behavioral variation among individuals has often been reported. However, stable among‐individual differences along a behavioral continuum—reflective of personality variation—have only recently become a key target of research. While a vast body of descriptive literature exists on animal personality, hypothesis‐driven quantitative studies are largely deficient. One of the main constraints to advance the field is the lack of suitable model organisms. Here, we explore whether N. furzeri could be a valuable model to bridge descriptive and hypothesis‐driven research to further unravel the causes, function and evolution of animal personality. As a first step toward this end, we perform a common garden laboratory experiment to examine if behavioral variation in the turquoise killifish Nothobranchius furzeri reflects personality divergence. Furthermore, we explore if multiple behavioral traits are correlated. We deliver “proof of principle” of personality variation among N. furzeri individuals in multiple behavioral traits. Because of the vast body of available genomic and physiological information, the well‐characterized ecological background and an exceptionally short life cycle, N. furzeri is an excellent model organism to further elucidate the causes and implications of behavioral variation in an eco‐evolutionary context.

shown to influence an array of ecological and evolutionary aspects, including distribution patterns, feeding niche, population growth and persistence, migration, dispersal, species interactions, social evolution, and adaptive potential (Briffa & Weiss, 2010;Mittelbach et al., 2014;Wolf & Weissing, 2012). For instance, Ioannou, Payne, and Krause (2008) illustrated that bolder three-spined stickleback (Gasterosteus aculeatus) preyed more heavily upon Chironomus prey, whereas Chapman et al. (2011) showed that bolder individuals of roach (Rutilus rutilus) had a higher tendency to migrate. The study of individual behavioral responses is, therefore, not only interesting from a fundamental perspective, but is also important for, for example, nature conservation, harvest and resource management as outlined in more detail by Mittelbach et al. (2014).
While a growing body of literature reports on animal personality in a variety of species, the ecological underpinnings that determine personality and its consequences remain poorly understood (Adriaenssens & Johnsson, 2013;Mittelbach et al., 2014). In addition, underlying proximate mechanisms (e.g. genetic, physiology) should be identified (Briffa, Sneddon, & Wilson, 2015;Oswald, Singer, & Robison, 2013) and animal personality should be studied across the ontogeny of organisms. This is a crucial step toward understanding the function, evolution, and mechanism of animal personality (Stamps & Groothuis, 2010). To tackle these goals, studies of individual behavioral variation in natural populations of well-characterized species along with laboratory and mesocosm studies have been promoted Mittelbach et al., 2014).
A number of fish model organisms are used in various fields of biological research . Best studied species include zebrafish (Danio rerio), medaka (Oryzias latipes), fathead minnow (Pimephales promelas), and stickleback (Gasterosteus aculeatus). One constraint with these models is their relatively slow maturation and long lifespan, hampering whole life or multigenerational studies (Harel et al., 2015). Annual killifish combine the perks of traditional fish models with the short generation time of invertebrate model species . Still, in order to be used as models in personality research, the existence of consistent amongindividual variation in behavior needs to be demonstrated. Species of the African genus Nothobranchius inhabit temporary freshwater pools and are adapted to the seasonal desiccation of their habitat by completing their life cycle in typically 3-4 weeks (Cellerino, Valenzano, & Reichard, 2015;Polačik et al., 2016). Moreover, annual fish produce drought-resistant, dormant eggs that form an egg bank in the sediment and hatch when the pool is inundated again (Reichard, Polačik, & Sedláček, 2009).
The African annual killifish Nothobranchius furzeri (Turquoise killifish) has the shortest lifespan of any vertebrate in captivity. The species typically matures after three weeks (both in laboratory and in field conditions) and survives for only 5-6 months post-hatching (Blažek, Polačik, & Reichard, 2013;Terzibasi et al., 2008), with even shorter lifespans recorded for inbred (homozygous) lines Wang, Promislow, & Kaeberlein, 2015). The "fastest" fish in our breeding facility even reached maturity in a little under two weeks (E. Thoré, Personal observation). In addition to a very short lifespan and generation time, N. furzeri is relatively easy to culture . Moreover, individuals typically have a high reproductive output and are naturally bold in behavior (Cellerino et al., 2015;Polačik et al., 2016). Dormant eggs can easily be stored and allow for a synchronized hatching for experiments (Philippe et al., 2017;Polačik et al., 2016).
In this study, we explore and discuss how N. furzeri could be a valuable model organism to move from descriptive studies to quantitative personality research to further unravel the causes, function, and evolution of animal personality. Despite the magnitude of recent advances in a range of biological disciplines that continue to promote N. furzeri as a promising model organism, individual behavioral variation in N. furzeri remains to be examined. Consequently, it remains unknown whether behavioral differences among N. furzeri individuals reflect personality variation or are due to random stochastic variation. We study a range of behavioral measures in N. furzeri individuals to determine whether variation is reflective of stable individual differences or rather represents random stochastic variation as a first and necessary step for quantitative personality research. In addition, we examine if and how different behavioral traits are correlated with each other. As personality variation has already been confirmed in a range of fish species (Budaev & Brown, 2011), we expect stable individual differences in behavioral expression and correlated behaviors also to be present in N. furzeri.

| General setup and fish maintenance
A total of 20 N. furzeri fish (9 females, 11 males) were reared from egg to adulthood while quantifying a range of behavioral measures under common garden rearing conditions in the laboratory.
Fish originate from the natural population MZCS-414 (central Mozambique) and had been laboratory-reared for three generations under optimal common garden conditions prior to the start of the experiment in accordance with the protocols as specified by Polačik et al. (2016). Fish were hatched (synchronized) by submerging eggs and peat in reconstituted water (type II RO water with added Instant Ocean salt mix; 8.3 pH, 600 μS/cm conductivity) at a temperature of 14°C, based on the protocol outlined by Polačik et al. (2016). Two days post-hatching, fish larvae were transferred to housing tanks at a density of 20 larvae per 4-L water. At an age of 2 weeks post-hatching, fish were transferred to 10-L aquaria in groups of 10. Three weeks post-hatching and for the remainder of the experiment, fish were housed individually in 9-L tanks for individual monitoring. Fish were placed in a housing compartment (one compartment per tank, approx. 12 cm L × 19 cm W × 16 cm H) to habituate them to the tank setup for behavioral testing (see below). Each compartment was provided with an air-driven filter to ensure good water quality. The sides of the tanks were covered with opaque plastic partitions to prevent confounding social contact between individuals. If social contact was allowed, neighboring dominant males could have interacted more with each other and have higher energetic needs than would neighboring females or submissive males. Water was renewed every 2 days from the moment of hatching until 3 weeks posthatching. Afterward, water was renewed on a weekly basis. Fish larvae were fed an ad libitum quantity of Artemia franciscana nauplii (Ocean Nutrition, Essen, Belgium) twice a day until 3 weeks post-hatching and frozen Chironomus larvae (Ocean Nutrition, Essen, Belgium), supplemented with Artemia nauplii, from an age of 3 weeks onward. On observation days, fish were fed only once a day to avoid interference with the observations. At an age of 6 weeks, a small amount of the solvent DMSO (dimethyl sulfoxide) was added to the water of the housing tanks as part of a concurrent experiment studying the effects of antidepressant exposure on behavioral expression in killifish. The applied solvent concentration was, however, negligible (0.00001 vol%) and identical for all individuals. Still, to account for any potential differences before/after DMSO addition, we included it as a random factor in our statistical analyses.

| Behavioral setup
At an age of one month post-hatching, individual fish were subjected to four different behavioral tests, each of which was repeated once per week for a total of five consecutive trials per test. For each trial, each fish was placed individually in an experimental arena and allowed to acclimate for 5 min prior to the start of the observation trial. After each behavioral trial, individuals were transferred back to their housing tanks. Behavioral tests included (a) an emergence test; (b) an open field test. To also include behavior with direct ecological relevance, a habitat choice test (c) and a life skills test (d) were included. For practical reasons and to minimize behavioral changes associated with timing of the day (Tran & Gerlai, 2013), each sampling burst per behavioral trial lasted for a maximum of 3.5 hr. Fish were divided in two observation cohorts for practical reasons, following the scheme presented in Table 1. In order to motivate fish to explore the arena and to prevent disinterest in the applied food, fish were abstained from food for 24 hr before the emergence test and life skills test.
As body size can be an important covariate of behavioral expression (Polverino, Bierbach, Killen, Uusi-Heikkilä, & Arlinghaus, 2016), it was determined on the day before the first behavioral observations by briefly transferring each individual to a petri dish with a small amount of water and taking size-calibrated photographs (mean ± SD = 23.94 ± 1.70 mm). Measurements were performed using the open source image processing software ImageJ 1.50i (Schneider, Rasband, & Eliceiri, 2012).

| Emergence test
Exploration tendency was studied by means of an emergence test.
Individuals were transferred to an experimental arena (Figure 1a), similar to their housing tank (i.e. a smaller compartment, separated from a "novel" larger compartment). After the acclimatization period, a doorway was opened to allow the individual to leave the small compartment and explore the larger one. Latency time to enter the newly available compartment was recorded during the next 45 min.
Fish that did not enter the compartment during this time were assigned the maximum score of 45 min.

| Open field test
An open field test was performed to explore basic locomotor activity parameters and the propensity to take risks (boldness). Individuals

| Habitat choice test
The test arena to explore habitat choice was divided in two equal was recorded for 30 min after 5 min of acclimatization.

| Life skills test
Feeding and antipredator behaviors were explored by means of a life skills test (Figure 1d). The experimental arena was virtually divided in four equal-sized zones. After an acclimatization period, the test was initiated as soon as fish entered either zone 1 or 4, after which food (Chironomus larvae) was gently added in zone 3. The latency time to feed was assessed. At the onset of feeding, an avian predator attack was simulated by means of a suspended, weighted 15-ml falcon tube (beak shaped, opaque) that was dropped and allowed to touch the water surface in zone 3 (see for instance Sih, 2007 andHedgespeth, Nilsson, &Berglund, 2016 for a similar setup). Subsequently, the time until a fish that froze or fled in response to the simulated attack resumed movement and the time that was needed to resume feeding were recorded. Again, a maximum time of 45 min was allowed.
All behavioral measures were recorded (top view) using Logitech C920 HD Pro webcams and were manually analyzed afterward (observer-blind), except for open field data which were analyzed using EthoVision XT Version 9.0 video-tracking software (Noldus Information Technologies Inc; www.noldus.com). for analysis in the current study). This approach enabled us to reduce the use of laboratory animals, by assessing multiple research objectives per experiment and avoiding the subsequent use of additional animals consistent with the "three Rs" guiding principles for more ethical use of laboratory animals (Fenwick, Griffin, & Gauthier, 2009).

| Statistical analysis
All statistical analyses were conducted in R 3. to account for behavioral changes over time) as fixed factors, including interaction between sex and body size and between sex and trial number. Fish identity was added to the model as random factor. To also account for variation explained by potential differences between cohorts and before/after DMSO treatment (0.00001 vol%), these factors were added to the models as random factors. Significance of fixed effects was tested using Wald chi-square tests (car package; Fox et al., 2017). To determine if individual behavioral variation is reflective of personality variation, repeatability measures were calculated using the rptR package (Stoffel, Nakagawa, & Schielzeth, 2018) as the between-individual variance over the sum of between-individual and residual variance (Nakagawa & Schielzeth, 2010

| Behavioral response variables
Latency time (in seconds) to enter the novel environment was assessed in the emergence test and was log-transformed to meet model assumptions. Due to a low resolution in the data (36% did not emerge within the given 45 min and were assigned the maximum value), model assumption of homoscedasticity could not completely be met. Therefore, these results should be interpreted with some caution.   the life skills test, latency time to feed before attack, latency time to resume feeding and time till movement after attack were repeatable with R = 0.174 (p = 0.012), R = 0.108 (p = 0.011), and R = 0.32 (p = 0.005), respectively. Mean value, standard deviation and minimum and maximum value for all behavioral response variables are presented in Table 2. Results of the mixed models are presented in Table 3. Correlation coefficients per pair of behavioral traits are presented in Figure 2.

| D ISCUSS I ON
Individual behavioral variation is a common phenomenon in animals. Scientists now increasingly recognize the importance of decomposing such variation in among-and within-individual variation and argue that consistent individual differences along a behavioral continuum exist (Briffa & Weiss, 2010;Roche et al., 2016;Sih et al., 2004). Here, we investigated if behavioral variation in the upcoming vertebrate model organism N. furzeri is reflective of stable individual differences or rather represents random stochastic variation.
Overall, our results support consistent inter-individual differences in multiple behavioral responses, suggestive of personality variation.
All behavioral measures that reflect locomotor activity and the propensity to take risks were found to be repeatable. This suggests that consistent behavioral differences exist among N. furzeri individuals.
Locomotor activity and movement resumption after a simulated predator attack of N. furzeri individuals differed between the trials.
This variation should most likely be interpreted as the result of behavioral plasticity or age-related behavioral changes. However, our results show that relative differences in both locomotor activity and movement resumption after a simulated predator attack between individuals were maintained across trials. This implies the existence of consistent variation in behavior among individuals. Since behavioral responses and body size were not correlated in the studied fish we suggest that neurophysiological rather than biophysical differences underlie these observations, which is consistent with findings on zebrafish (Danio rerio) (Tran & Gerlai, 2013).
In personality research, behavioral traits are often found to be dependent of each other (Class & Brommer, 2015). Such behavioral correlations are however not unanimously supported in the literature (Garamszegi & Herczeg, 2012). Likewise, in this study we found only remark however that the current study is not optimally designed to explore behavioral correlations, as such studies typically require larger sample sizes (Garamszegi & Herczeg, 2012).
The historical lack of interest in between-individual behavioral variation has been used as justification to launch the field of personality research (Beekman & Jordan, 2017). However, this notion recently received criticism. It can be argued that individual behavioral variation was already the implicit corner-stone of behavioral ecology-a discipline that explains animal behavior from a functional and evolutionary perspective (Beekman & Jordan, 2017;Owens, 2006). Moreover, the bulk of animal personality literature is descriptive rather than hypothesis-driven and often lacks an evolutionary context and insight into the mechanistic underpinnings of behavioral variation (Beekman & Jordan, 2017;Roche et al., 2016;Sih, 2013).
In light of the task ahead, the use of model organisms could be indispensable to bridge descriptive and hypothesis-driven research, as suggested by Owens (2006). Behavioral ecologists typically study behavior of wild organisms rather than that of model organisms (Monaghan, 2014;Owens, 2006). Owens (2006) (Alfred & Baldwin, 2015;Owens, 2006;Parichy, 2015). Consequently, behavioral ecologists miss out on a large amount of biochemical and physiological data that is available for traditional model organisms and fail to effectively link individual behavioral variation to the genetic and physiological mechanisms that underpin this variation (Beekman & Jordan, 2017;Owens, 2006). Although (field) studies on natural populations are of crucial importance, model organisms are under-used in behavioral ecology. This has, in our opinion, contributed to the predominantly descriptive nature of animal personality research. The use of model organisms in combination with natural populations could extend the scope from descriptive studies to targeted quantitative research designs. In addition, it would also solve the lack of replication across laboratories that is common in behavioral ecology and help to validate findings, adding to the scrutiny of the field (Owens, 2006).
Nothobranchius furzeri is already a commonly used model organism in a range of biomedical research fields, in ecology and in evolutionary biology . We argue that N. furzeri has great potential to further add to the advancement of ethology and behavioral ecology. Whereas personality variation has been reported in a range of traditional model organisms such as zebrafish (Tran & Gerlai, 2013), stickleback (Jolles, Taylor, & Manica, 2016), and rainbow trout (Elias, Thrower, & Nichols, 2018), this is the first study to report on personality variation in N. furzeri.
The main advantage of N. furzeri is its extremely short maturation time (3 weeks) and lifespan (typically <6 months). This is much shorter than that of traditional model organisms such as zebrafish with a typical maturation time of 2 months and lifespan of up to 5 years (Lawrence et al., 2012). In addition, the life history of the species is well characterized across ecological gradients (e.g. aridity, predation) and ongoing ecological and evolutionary research continuously adds to this knowledge Watters, 2009). This facilitates field studies to further elucidate the ecological causes and consequences of behavioral variation (Mittelbach et al., 2014). In complement with this ecological information, a whole brain atlas (D'angelo, 2013) along with an annotated genome (Reichwald et al., 2015;Valenzano et al., 2015) and transcriptome (Di Cicco et al., 2011) allow for an in-depth investigation of neurophysiological and genetic underpinnings of behavioral variation.
Moreover, successful transgenesis and the generation of transgenic lines (Hartmann & Englert, 2012) allows for the construction of specific lines. If such lines are characterized by different behavioral profiles, the molecular mechanisms underlying behavioral variation and the ecological and evolutionary implications of personality variation could be studied under well-controlled experimental conditions (Tran & Gerlai, 2013).
Whereas studying the development of animal personality across ontogeny is pivotal to our understanding of the proximal causation, function (adaptive value), and evolution of personality, this remains largely understudied. Understanding the development of animal personality typically requires time-consuming research as behavioral changes across the lifespan of individuals need to be monitored (Stamps & Groothuis, 2010). In this regard, the short lifespan of N. furzeri offers a major advantage over other vertebrate species.
Moreover, its fast life cycle allows for time-and cost-efficient examination of environmental underpinnings of behavioral variation across ontogeny-including experiential factors-and multigenerational setups.
While in the current study we identified stable individual differences in behavioral expression, thereby adding necessary fundamental information to the descriptive animal personality literature, we also provide a potential stepping stone to quantitative research designs by introducing the model organism N. furzeri. The inclusion of N. furzeri in behavioral sciences answers the call for an increased diversity in studied organisms (Monaghan, 2014). Its wellcharacterized biomedical and ecological background in combination with its short life cycle make it an excellent model organism to further elucidate the causes and implications of behavioral variation in an eco-evolutionary context.

ACK N OWLED G M ENTS
We are grateful to Dr. Martin Reichard and coworkers for providing the parental fish for this project. We furthermore thank the reviewers for their constructive comments on the manuscript.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
The study was designed by ESJT and TP and performed by ESJT and LS. Data was analyzed by ESJT. The manuscript was written by ESJT and reviewed by TP, LS, CP, and LB. All authors gave final approval for publication.