Animals and housing
Adult male Siamese fighting fish B. splendens (N = 56) were acquired from a commercial supplier, Grosvenor Tropicals (Lisburn, Northern Ireland). Fish were bred and kept in captivity, but housed individually by the supplier as juveniles and adults, which addresses issues related to the degree of familiarity between them and the effects this might have on competition [60, 61]. After acquisition, animals were again housed individually in tanks (32 cm length × 22 cm width × 31 cm height) and visually isolated from their neighbours, to prevent stress-effects from aggressive interactions during housing and further limit familiarity effects. Housing tanks were filled with 15 L of water, enriched with plants, toys and shelter. Tank water was filtered, heated, aerated, and regularly tested for chemical quality, twice-weekly changed and kept at 26 ± 1 °C, 7.2 ± 0.4 pH and a regulated bacterial cycle. Further, regular health checks were carried out for signs of common diseases (as listed by Monvises et al. [62]). Each fish was kept on a moderate level diet of commercial feed (Hikari© Bio-gold; 4 pellets twice a day) with 38% animal protein, considered ideal in terms of growth and conversion rates [63]. On the day of experiments, animals were fed only after the staged contests to standardise motivation [1]. Light conditions were controlled at 12 h photoperiods (0700–1900), with experiments conducted during light periods (300 lx luminosity and 350–600 nm wavelength at water surface).
Size measures
After a 72-h acclimation to laboratory conditions, measures of size were recorded for all fish to be used as estimates of RHP (see review by Arnott and Elwood [7]). First, the wet weight of each fish was calculated by the difference in mass between a water-filled container with and without the fish (cf Kareklas et al. [36]). Then fish were briefly removed from the container and placed with some water on a prepared waterproof surface marked by 1 cm squares. There, the anal and caudal fins of the fish were gently extended using a plastic pipette and fish were photographed using a suspended camera (Sony HDR CX190E handy-cam video camera) and immediately placed back into the water-filled container to be returned to their tank. Using Image-J, we used the 1 cm marked squares from the pictures as a reference for calibrating digital distance to actual distance (in cm) and used the software’s measurement tools for calculating standard body length (in cm: from tip of snout to the base of caudal peduncle) and the area of extended anal and caudal fins (cm2) for all fish.
Fin size and body length quantify morphological information that can be visually assessed in opponents during display [31], but each likely to provide separate information about an opponent’s RHP; length may designate overall size, but fin size may signal attack performance (e.g. charging and tail beating capacities [64]). Furthermore, preliminary analyses indicated that standard body-length and fin size (combined caudal and anal fin size) were only weakly related (r2 = 0.297; lack-of-fit: r2 = 0.572). Conversely, weight cannot be visually assessed but can provide a more inclusive measure of RHP because it can reflect the composite morphological state of animals, but may also reflect the private information fish have about their own physical capacities (i.e. their own perception of size or strength [22]). Indeed, weight strongly predicted standard length (r2 = 0.549; F56 = 65.76, P < 0.001) and combined fin size (r2 = 0.445; F56 = 43.31, P < 0.001) across all animals, focals and opponents. Although all animals were adults, weight may also reflect differences related to age and reproductive activity, but the relation here with length and fin size suggests that it at least represents morphological elements of RHP. Therefore, weight was used as an indicator of composite inter-contestant asymmetries in RHP and used for matching focal fish with comparatively bigger or smaller opponents, whereas length and fin size were considered as covarying factors that may predict added inter-individual size-related effects and visual-assessment biases. The mean relative weight difference of opponents compared to focals was 0.38 g ± 0.56 SD and ranged between − 0.64 g and + 1.63 g.
Experimental procedures
Focal fish (n = 30) were housed in their individual tanks for 2 weeks before experiments to ensure territorial establishment via the use of landmarks in their tank [51] and to allow time for bubble-nest building. The experiment consisted of four weekly contests so that focal fish faced both a bigger and a smaller opponent, each under both noise and control acoustic conditions (a within-individual 2 × 2 factorial design), with acoustic-treatment order being randomised and opponent-size order counterbalanced across individuals. Opponent size was manipulated by replacing opponents between contests, based on their relative weight. Acoustic conditions were manipulated using a soundproofed experimental lid with an embedded speaker (Fig. 2a), which either played white noise under treatment or remained silent during controls (see similar protocols of air suspended speakers eliciting response in related anabantoid fish [47, 65]). Focal and opponent tanks were kept with some distance between them (~ 2 cm) to control for sound vibration effects on opponents. White noise treatments resembled distant-traffic noise, low-pass filtered to 100 Hz frequency with 6 dB kHz− 1 decrease towards higher levels [66] (set to 80 dbA at 1 m, SL-100 Voltcraft, Hirschau meter). Underwater sound recordings were performed using a hydrophone (Aquarian Audio H2a-XLR; omnidirectional; sensitivity: -180 dB re. 1 V/μPa), placed centred on the bottom of a tank (conservative measure of greatest depth from source) and analysed in the Audacity® software. Compared to controls, the noise treatment exhibited markedly greater changes in sound pressure level (Fig. 2b) and five-fold increases in frequency (Fig. 2c), kept within the 5 kHz maximum hearing threshold attributed to B. splendens based on its auditory structure similarities to other anabantoid fish, such as Trichopsis spp [47, 48]. The maximum of ~ 40 dB change in sound pressure levels at frequencies of 1–4 kHz (Fig. 2d) under noise treatment reflected nominal low-magnitude underwater noise in areas near terrestrial city traffic or with light shipping [49, 67] and matched profiles eliciting response in other gouramis (maxima of 5 kHz; 20–40 dB changes [47, 68, 69]). The lid and noise conditions were experienced by focal animals for 10 min before contests, allowing for both acclimation to the lid and assessment of noise conditions, and maintained only for the duration of each staged contest. The lid of housing tanks was opaque and maintained on opponent tanks during contests to standardise lighting conditions after placing the soundproofed experimental lid.
Before each contest, the water surface of tanks was photographed for recording bubble nests on the water surface. Contests were staged in the form of simulated territorial intrusions by removing opaque covers that visually isolated stimulus fish (n = 26) in immediately neighbouring tanks (acclimated overnight). The onset of contests was set at the first agonistic behaviour by focal fish, either in terms of bite attacks, tail beats, or display (either frontal display - presenting extended gills to opponent- or lateral display - presenting side with flared fins to opponent [31]). Fight motivation was probed by measuring focal recovery times from startles elicited by dropping a glass marble (24 g) through one of the two bilateral tubes in the lid, with the side counterbalanced across individuals and consecutive contests, and between within-contest consecutive probes. The marble landed in the water surface from a 10 cm height and invoked a startle response in focal animals via visual and mechanosensory cues from the distinct splash (Fig. 2a [36,37,38,39];). This probe was performed twice during contests, at 5 min after onset of contests and again at 5 min after focal fish resumed interaction. Following recovery from the second startle, fish were allowed to interact for a final 5 min period before concluding the test by replacing the opaque cover between tanks. Although the overall duration of the staged events varied due to differences in startle durations, the initial 5 min interaction and the two 5 min interactions after each probe ensured that all contests had a standard 15 min interaction between focals and opponents. Contests were video recorded (Sony HDR CX190E handy-cam video camera) and the experimenter remained hidden during test-related manipulations by a large cover (150 cm × 150 cm) and away in interim periods, to minimise interference.
Behavioural measures
From the pictures taken before each contest, we first recorded the presence of bubble nests and then measured nest size (in mm2) using the Image-J software to calibrate digital to actual area. Nest-presence was identified in overall 69 contests: 37 out of 60 under control conditions (19 against bigger opponents, 18 against smaller opponents) and 32 out of 60 under noise treatment (18 against bigger opponent, 14 against smaller opponents). This suggests that fish had at least some underlying differences in reproductive state, but our within-individual experimental design controlled for the influence of this on variations between contests. Further, there were no temporal effects (Binary; ID × Day: χ23, 119 = 0.31, P = 0.957) or noise effects (ID × Noise: χ23, 119 = 0.10, P = 0.758) on individual nest-presence, indicating that the tendency of individuals to construct nests or not was consistent across the experimental period and independently of acoustic condition. This tendency could be attributed to other factors, which according to evidence could include stressors such as temperature or water quality [70, 71], but these parameters were controlled and standardised in our study. In addition, alternative effects from individual differences in baseline stress levels are unlikely to have a direct impact, but could influence the relationship between body size and nest quality [70, 71]. However, mean weight did not differ between males that tended to build nests and those that did not (t = 0.66, P = 0.513), and there was no significant relationship between nest size and body size (length: r = 0.088, P = 0.340; weight: r = 0.081, P = 0.381). On the contrary, the size of nests exhibited individual differences over time (ID × Day: χ23, 119 = 444.38, P < 0.001). We therefore, used the size of nests as an indicator of individual changes in construction across contests, with bubble nest size before each contest used to calculate differences in construction between weeks (± mm2), for examining changes in territory use over time. Because the absence of nests was set at 0 mm2, the variation reflected changes driven by fish who tended to construct nests.
From the recordings of contests, the time taken to recommence display following each of the two startle probes was used as an estimate of aggressive motivation, with shorter recovery times (in seconds) indicating greater aggressive motivation [36,37,38,39]. The combined duration of frontal and lateral display across the contest was also measured, both for focals and opponents. Overall, fish exhibited display across all contests and total display duration was significantly related to weight (r = 0.310, P < 0.01), indicating that weight was a robust predictor of RHP. In addition, opponent display duration was not affected by the acoustic conditions in the focal tank (t1,119 = − 0.22, P = 0.826), which suggests that noise was largely contained to the focal tank and did not affect the behaviour of opponents during contests. The startle durations of focals were strongly predicted by display durations (R2 = 0.396, F1, 119 = 77.43, P < 0.001), confirming their representation of aggressive motivation. Although startle durations increased (t1,119 = 5.90, P < 0.001) between the first (7.52 s ± 7.98 SD) and second probe (26.49 s ± 39.88 SD), a likely effect of energetic expenditure [7, 72], variation in startle duration was consistent between probes (Pearson, r = 0.650, P < 0.001) indicating consistency in inter-individual differences. Therefore, the mean startle duration of the two probes from each contest was used as a measure of individual aggressive motivation in statistical tests.
Statistical analysis
Data calculations, statistical tests and graphical representations were carried out in the statistical software Minitab® version 17 (Minitab Inc., State College, PA, USA) and SPSS version 22 (IBM Corp., Armonk, NY. USA). Normality was tested for both raw data and residuals from the models (Shapiro-Wilk) to inform the choice and fitness of statistical tests. Across tests, Cohen’s d was used as a measure of effect size, based on the mean and standard deviation, and R2 as an effect size measure for interactions with temporal changes.
According to our first two hypotheses, a decrease in territory value under noise is expected to reduce defence motivation (1) when facing bigger opponents that pose higher costs but (2) not when a nest is present, because it increases territory benefits. Therefore, we first tested whether mean startle durations per contest, our measure of defence motivation, was affected by the interaction of acoustic condition (noise playback or control) with relative opponent RHP (bigger or smaller in weight than focal) and nest presence. For this a mixed effects model with a log-link function was used because startle duration data was not normally distributed (P < 0.01), but residuals from the model conformed to normality (P > 0.05). The model included opponent length and fin size as covariates to control for assessment biases from visual signals [36, 39], as well as day and individual identity of focals and opponents as random factors to respectively control for carry-over effects and pseudoreplication. Post-hoc pairwise comparisons (Fisher’s LSD) were then performed to examine differences between contests under different conditions.
According to our third hypothesis, (3) a decrease in territory value under noise is expected to also reduce subsequent motivation to use the territory for nest-building, which may include interaction effects with opponent RHP, where fighting relatively bigger opponents imposes added costs. As such, we tested the effects of acoustic conditions and relative opponent size on both temporal variation in bubble-nest size across contests, to identify carry-over effects from sequential changes [73], and immediate nest-size changes following each contest. For effects on the temporal nest-size variation, a Gamma model with a log-link function was used on the bubble-nest size data to address their skewness towards zero due to the high no-nest incidence (\( {\overset{\sim }{\mu}}_3 \) > 3.69) and their consequent lack of normality (P < 0.005). Recent evidence [74] shows that skewness violations do not significantly contribute to estimate violation in mixed-type models, especially when data are not severely skewed \( \Big({\overset{\sim }{\mu}}_3 \) < 3). The residuals from the model were in line with this by exhibiting notable decreases in skewness (\( {\overset{\sim }{\mu}}_3 \) = 1.77) and increases in normality (P = 0.01). This model was used to test for variation over time (days in housing tank) and whether this was affected by acoustic treatment order and its interaction with opponent-size order across contests, where focal length was included as a covariate (to control for body-size effects on bubble-nest construction [28, 70, 71];) and individual identity was included as a random factor (to control for pseudoreplication). For the immediate effects, we pooled calculated changes in nest size for the week following each contest (normally distributed) and used an ANOVA model to test whether these changes relied on acoustic condition (noise playback or control) and relative opponent RHP (bigger or smaller in weight than focal) during the contest, including post hoc one-sample t-tests for assessing significant differences from no change under different conditions.