When animals live in long-term groups, the potential for conflict is high. Conflict is costly, so an individual’s decision to engage depends on the information it has about the costs and benefits of fighting. One source of information could be past contest experience, where previous winners/losers typically become more likely to win/lose in the future. However, repeated interactions can familiarize individuals with conflict and provide opportunities to learn to become better fighters, regardless of outcome. We explored how individuals integrate information from previous contests to inform future encounters in a group-living fish, Neolamprologus pulcher. We gave contestants single, reinforcing, and contradictory experiences and measured behavior and post-fight water-borne levels of androgenic steroids (testosterone, 11-ketotestosterone). Contradictory outcomes were associated with reduced investment in fighting. More fighting experience did not lead to greater investment in fighting, as consecutive losses resulted in reduced aggression. Also, there was no effect of fighting treatment on water-borne androgen concentrations. Interestingly, there were sex differences in which behaviors were influenced by experience, and in whether body mass was associated with androgen concentrations, which could indicate that males and females vary in how perceived fighting ability changes with contest experience. Our data reveal the complex ways in which repeated experiences can alter an individual’s propensity to invest in conflict. Repeated interactions associated with predictable changes in behavior can contribute to rank stability in groups and our results indicate that whether and how they do depend on the quality and quantity of interactions plus individual factors such as sex.

Neolamprologus pulcher were purchased from Old World Exotic Fish, Inc. in Miami, FL in December 2018, or were F1 offspring of these fish. Contestants were housed in well-established social groups that maintained consistent group membership for at least two years prior to the start of this experiment. No contestant/opponent pair had been housed together or in adjacent tanks for at least two years prior to this experiment to reduce the effects of familiarity. Two weeks prior to the first contest trial, each fish was measured for total length (TL, to the nearest 0.01 mm), standard length (SL, to the nearest 0.01 mm), and mass (to the nearest 0.001 g). Each fish also was sexed by examination of the external genital papilla (Stiver et al. 2005) and received a unique visible implant elastomer tag (a subcutaneous injection of a silicone-based colored liquid) (Jungwirth et al. 2019) at this time. Only sexually mature animals (more than 30 mm in length) (Hellmann, Ligocki, et al. 2015; Antunes and Taborsky 2020) were used in this experiment. Individuals were randomly assigned ‘focal’ or ‘opponent’ status. A total of 43 contestants and 72 opponents were used for this experiment. Focal fish occupied a range of social ranks in their home tanks (n = 10 breeder females, 10 helper females, 11 breeder males, 12 helper males). Body size is a strong predictor of contest outcomes in N.pulcher (Reddon et al. 2011), so same-sex opponents were haphazardly selected to be 8.05 ± 2.53 mm smaller or larger in absolute SL than the focal fish to create a winning or losing experience, respectively. All focal fish were naïve to experimental procedures. If an opponent was to be used in more than one trial (n=24 opponents used in more than one trial), they had at least one week between trials (Miklósi et al. 1997). Focal fish sometimes were used as opponents (n=6) after they experienced contest trials as the focal contestant; at least one week elapsed between trials before a focal fish served as an opponent.

Contest trials

Contests took place November - December 2021. Focal fish were randomly allocated to one of six experimental treatments; a single contest (win, lose), reinforcing contest experiences (win-win, lose-lose), or contradictory contest experiences (win-lose, lose-win). All contests were conducted between same-sex pairs prior to feeding in experimental tanks detailed in Figure 1, similar to (Reddon et al. 2011; Hick et al. 2014). Contestants competed for access to a single terracotta flowerpot placed in the middle of the experimental tank, which we refer to as the ‘central contest zone.’ N.pulcher use these pots for shelter and breeding in our laboratory and will fight for access to this resource (Reddon et al. 2011; Hick et al. 2014; Pisanski et al. 2015). Contestants that experienced more than one contest were immediately moved from their first contest to a 30-minute acclimation period for their second contest with a different opponent. Experimental tanks were sanitized with ethanol and the water was replaced between each trial. 

Mirror tests are an appropriate proxy for aggression toward live opponents in N.pulcher (see Electronic Supplementary Material A, (Balzarini et al. 2014)). After their final contest (after the second contest for individuals that received a reinforcing or contradictory experience and after the only contest for individuals that received a single experience), contestants had a 30-minute acclimation period before a 20-minute mirror test. Mirror tests took place in an independent arena set up in the same way as the experimental tanks but with a mirror attached to a randomly selected barrier that was not removed and gave contestants access to 2/3 of the tank.

Behavioral observations

A high-definition camcorder was used to record each contest and mirror test.  The number of aggressive behaviors performed by the focal contestant and duration of time the focal fish spent in the central contest zone were scored continuously using Behavioral Observation Research Interaction Software (BORIS) (Table B.1).  Displayed acts of aggression such as spreading of the fins or swimming parallel to the mirror or opponent were scored as restrained aggression whereas aggressive acts that involved contact with the mirror or opponent were scored as overt aggression (see Electronic Supplementary Material B). The scorer was blind to experimental treatment during scoring.

All contest trials were observed live from behind a blind. Contests lasted for 20 minutes or until a ‘winner’ and a ‘loser’ could be determined. A contestant was declared a ‘loser’ when it displayed three consecutive submissive behaviors toward its opponent, or when it held a submissive posture while avoiding the other fish (Reddon et al. 2011; Hick et al. 2014). A contestant was declared a ‘winner’ when it chased its opponent three consecutive times or when it prevented its opponent from approaching the central shelter (Riebli et al. 2011; Dijkstra et al. 2012). When one fish was declared a ‘winner,’ its opponent was automatically considered a ‘loser’ and vice versa. Contests were terminated as soon as a ‘winner’ and a ‘loser’ were determined. If there was no clear winner or loser at the end of the 20-minute observation period, then the contest was declared a tie. Any focal individual that experienced a tie or did not experience their assigned contest outcome were returned to their home tank without completing a mirror trial. 

Hormone measurements

Due to the small size of N.pulcher, it is not possible to consistently collect a large enough blood sample to measure steroid hormone concentrations without sacrificing the individuals. Therefore, steroid hormones were extracted from fish holding water so these individuals can be used for future studies and to provide a large enough sample volume to measure multiple hormones from a single sample (Bender et al. 2006; Sebire et al. 2007). This method extracts free, unconjugated hormones that have diffused from the bloodstream, through the gills, into the holding water and reliably reflect circulating hormone levels for N.pulcher (Scott et al. 2008). With passive diffusion, there is a time lag between maximum release rates reflected in water-borne hormone samples and maximum circulating hormone concentrations. Therefore, hormones were extracted two days after the focal fish experienced the contest trials.  

Fish were placed in individual beakers with 300 ml of distilled water in separate boxes for exactly one hour before they were returned to their original tanks. This water was kept at the same pH and temperature as water in the fish tanks, but no fish were housed in it prior to hormone extractions. Fish were exposed to this procedure three times prior to their contest trials to habituate them to the process so the procedure itself (e.g., transferring from original tank to beaker) would not affect hormone release rates (Ligocki et al. 2015). After this time, the holding water was passed through Grade 1 filter paper to remove any solids before being drawn through Sep-Pakâ solid-phase C18 extraction cartridges (500 mg sorbent, 3 ml cartridge; Watersä Corporation, Milford, MA) using a vacuum manifold. These samples were frozen at -80oC and shipped to the University of Alabama. Salts were purged from the columns with 2 ml distilled water, and free, unconjugated steroids were eluted with 3 x 2 ml of ethyl acetate into 13 x 100 mm borosilicate vials. A gentle stream of ultra-pure nitrogen (~7 bar) was passed over the samples in a 37ºC water bath to facilitate evaporation of the solvent, leaving a hormone residue. This residue was resuspended in 25 µl EtOH and vortexed for 1 min before adding 475 µl EIA buffer and vortexing again for 1 min (500 µl total resuspension volume).  A pooled sample of hormone was generated by combining 30µl from each experimental sample, representing the “average” water-borne hormone concentration across all animals. 500 µl of this pool was then serially diluted 1:1 to 1:128 with EIA buffer and assayed using enzyme-linked immunosorbent assays (ELISA) kits from Cayman Chemical (Ann Arbor, Michigan) for 11-ketotestosterone (11KT) and testosterone (T). Comparisons of the serial dilution curves to the standard curves indicated that the best dilution factor (i.e., that which was most likely to position samples on the linear phase of the standard curve) was 1:4 for 11KT and 1:50 for T. These dilutions were achieved by combining 100 µl of each experimental sample with 300 µl of EIA buffer (for 11KT) and 40 µl of each experimental sample with 1.96 ml EIA buffer (for T), followed by vortexing. Serial dilution curves also were parallel to the standard curve (11KT: t12 = 0.075, P = 0.94; T: t12 = 0.039, P = 0.97), validating the use of these ELISAs for N. pulcher. Assays were conducted on two 96-well plates per hormone with intra-assay coefficients of variation being 6.9% and 4.2% for T, and 3.7% and 4.5% for 11KT. Inter-assay coefficients of variation were 5.3% for T and 3.8% for 11KT.

Statistical analysis

Generalized linear models (GLMs) were used to measure associations between steroid hormone concentrations, behavioral response, and experimental treatment. Counts of restrained and overt aggression, duration of time spent in the central contest zone, and concentrations of T and 11KT were analyzed in five separate GLMs. For behavior measures, fixed effects of the full models included the three-way interaction between sex, type of experience (single, reinforcing, contradictory), and most recent experience (win or loss) plus focal animal mass as a covariate.

Although previous studies on N.pulcher found individuals of different sizes release similar amounts of steroid hormones over the same time interval (Bender et al. 2006; Bender et al. 2008; Scott et al. 2008), it is possible that hormone release rates may vary depending on the size of the individual (Sadoul and Geffroy 2019). To account for the potential effect of body size on hormone release rates, fixed effects of the full models for hormone concentration response variables included the three-way interaction between sex, type of experience, and most recent experience plus the two-way interaction of each of these variables with mass.

The best fitting model for each response variable was selected by likelihood ratio test using the step function in R. After running the generalized linear model, a Tukey post-hoc test was used to compare treatment effects. All statistical analyses were performed using R 4.1.1. We used the car and emmeans statistical packages (Fox et al. 2022; Lenth et al. 2022) and RStudio. The significance level for all tests was α = 0.05.

All statistical analyses were performed using R 4.1.1. We used the car and emmeans statistical packages and RStudio.