Conspicuous ornaments are often considered a result of evolution by sexual selection. According to the social selection hypothesis, such conspicuous traits may also evolve as badges of status associated with increased boldness or aggression towards conspecifics in conflicts about ecological resources. This study tested predictions from the social selection hypothesis to explain evolution of conspicuous red colour of the pelvic spines of the three-spine stickleback (Gasterosteus aculeatus). Wild non-reproducing sticklebacks were presented to pairs of dummies which differed at their pelvic spines, having either (i) normal sized grey or red pelvic spines, or (ii) normal sized grey or large red pelvic spines. The experimental tank was illuminated by white or green light, since green light impedes the sticklebacks’ ability to detect red colour. The dummies moved slowly around in circles at each end of the experimental tank. We quantified the parameters (i) which of the two dummies was visited first, (ii) time taken before the first visit to a dummy, (iii) distribution of the focal sticklebacks in the two zones close to each of the two dummies and in the neutral zone of the tank, (iv) close to which of the two dummies did the focal fish eat its first food-piece, and (v) time spent until the first piece of food was eaten. This was carried out for 22 females and 29 males sticklebacks. The results suggested no effect of the colour or size of the dummies’ pelvic spines, on none of the five behavioural parameters. Moreover, neither the colour of the pelvic spines of the focal sticklebacks themselves (as opposed to redness of the dummies’ spines) or their body length was associated with behaviour towards the dummies. Thus, this study did not support predictions from the social selection hypothesis to explain evolution of red pelvic spines in sticklebacks.

Wild sticklebacks were collected from the 140 m long and 50 m wide and landlocked freshwater Lake Pallvatnet, located at an altitude of 140 m at 67°31’N, 14°40’E in Bodø, North Norway. Fish were caught by traps made of 1.5 L soda bottles deployed along the shore at 0.2 – 1.0 m depth. The traps fished for 24 hours 9 – 10 September 2019. Captured sticklebacks were transported to Mørkvedbukta Research Station in Bodø and kept in a 80 l storage tank with continuously flowing water until further handling. The fish in the storage tank were daily fed frozen Chironomidae larvae (Akvarie Teknik, DeLang and Ekman AB, Filipstad, Sweden).

The experimental trials were carried out 13 September to 4 November 2019. Trials were run with a total of 37 males and 23 females. Twenty-four hours before each trial, one focal stickleback was isolated in a transparent Plexiglass tube (7.5 cm in diameter, 35 cm height), hereafter termed the “isolation tube”, and placed inside the storage tank with the other fish. The focal specimen could see its conspecifics in the storage tank and there was a continuous flow of water between the storage tank and the isolation tube. However, the isolated focal stickleback did not have access to food during these 24 hours in order to increase its motivation to forage during the upcoming trials. After 24 hours in the isolation tube the focal fish was transferred to an experimental tank which consisted of several devices as shown in Fig. 1. Two electric engines moved two dummies in circles in each end of the experimental glass tank. The circumference of the circles was 66 cm, the speed of the dummies was 2.7 cm s-1 which means that the dummies spent 24 s per lap. Two of the three different dummies (see below) were presented simultaneously in one trial. W.J. Rowland kindly molded the dummies in epoxy in 2003 as described in Rowland (1979). A 52.0 mm non-gravid female stickleback was used as model. The dorsal part of the dummies was painted black (85 Coal Black Satin, AAA0655, Humbrol Enamel, Kent, UK) and the ventral part grey (64 Light Grey Matt, AA0713). We mounted artificial pelvic spines on three different dummies. Artificial pelvic spines of approximately normal length (10 mm) were mounted on two dummies, whereas larger (20 mm) artificial pelvic spines were mounted on the third dummy. On one of the dummies with small artificial pelvic spines and the dummy with large pelvic spines, part of the pelvic spines was painted red (60 Scarlet Matt, AAA0655). The red was painted along the entire spine from the base to the tip, covering about 180° of the pelvic spines with the red part directed towards the “body” of the dummies. The remaining part of the pelvic spine on these two dummies was painted with the same grey colour as the ventral part of the body of the dummies (see above). The pelvic spines of the remaining third dummy (with normal length spines) were painted grey all over. The artificial spines were installed spread out away from the body. This leaves us with three different dummies: one with grey pelvic spines of normal length (“Normal-grey” abbreviated “NG”), one with red pelvic spines and normal length (“Normal-red” abbreviated “NR”), and one with red and large spines (“Large-red” abbreviated “LR”). Two Chironomidae larvae were presented in each of two Petri dishes below each of the two dummies. The experimental tank was illuminated by either white or green light. The white light came from one light bulb (Anslut, E27 2.5 W 140 lm, rendering average (Ra) of 80, article number 421433 at www.jula.no), whereas green light came from three bulbs (Anslut, E27 0.7 W 30 lm, article number 420698). Light intensity was 280 and 120 lux for white and green light, respectively, measured by a light meter (Amprobe LM-120 Light Meter, Glottental, Germany) inside the experimental tank where the dummies were located. Three sides of the experimental tank were covered by non-transparent green plastic foliage, whereas video-recordings were carried out through the fourth uncovered tank wall.  

The behaviour of each fish was studied under four different combinations of pair of dummies and colour of light. The combinations were (i) “Normal-grey” and “Normal-Red” dummies illuminated by white light (ii) “Normal-grey” and “Normal-red” dummies illuminated by green light, (iii) “Normal-grey” and “Large-red” dummies illuminated by white light (iv) “Normal-grey” and “Large-red” dummies illuminated by green light. Green illumination prevents the sticklebacks from using their red cues which impedes their ability to see red colours (Milinski and Bakker, 1990). Each of the four combinations of dummies and light-colour is hereafter termed a “subtrial”, whereas the combination of the four subtrials with each individual focal fish is termed a “trial”. We randomized both the sequence of the four subtrials with each fish, and to which of the two zones (zone “A” or “B”, Fig. 1) each of the two dummies appeared in each subtrial.

Five minutes before the actual start of a trial, the focal fish was transferred to the transparent Plexiglas tube (TPT) located inside the experimental tank (Fig. 1), in order to become familiar with the two dummies and the tank. A subtrial (the first subtrial of four in a trial) started by removing the Plexiglass tube by pulling a thread while hiding behind a tarpaulin to avoid disturbing the focal fish. The stickleback could then swim freely around in the “Neutral zone” or closer to the two dummies in “Zone A” and “Zone B”, and feed freely on Chironomidae larvae beneath any of the dummies (Fig. 1). Each subtrial was recorded by a GoPro (San Mateo, US) camera mounted on a tripod in front of the experimental tank. Between subtrials, the focal fish was kept in the isolation tube (in the storage tank) while preparing for the next subtrial with the same focal fish. The trials were carried out in a quiet room. Care was taken not to disturb the fish during the trials and the focal fish saw no humans from being placed in the TPT-tank until the end of the recordings in each subtrial.

After the termination of all four subtrials of a trial, the specimen used in this experiment was killed by an overdose of MS-222 and frozen in –20°C in darkness awaiting further analysis. Each individual was measured for total length to the nearest mm and weight to the nearest 0.001 g, and the sex was determined by inspection of the gonads. The ventral part of each fish was photographed using an Olympus E-M10 with a M.Zuiko ED 60 mm 1:2.8 macro lens and a Nissin i40 flash, with its pelvic spines erected and together with a standardized colour palette, for later quantification of the redness of the pelvic spines (see below). The water in the experimental tank was removed and replaced with fresh water before a new trial started with another focal stickleback. Intensity of the red colour (IR) of the pelvic spines of each stickleback and the reddish part of the colour palette were quantified separately from the digital photos in RGB mode by Adobe photoshop version 13.1. This method has previously been applied to quantify colour by several authors (Skarstein and Folstad, 1996, Skarstein et al., 2005, Villafuerte and Negro, 1998, Nordeide, 2002, Nordeide et al., 2006, Nordeide et al., 2008, Amundsen et al., 2015). In the quantifications, we encircled the pelvic spines and the reddish part of the colour palette and estimated the average density values for all three primary colours R, G, B (red, green, and blue) from the pixels enclosed in each of the areas (Villafuerte & Negro 1998). Intensity of red (IR) of the two pelvic spines and of the reddish cardboard was calculated as:  (IR) = (R/(R+G+B). The average IR-value for both the pelvic spines was used in the analyses, after correcting for differences between photos using the IR-value from the reddish cardboard (see Nordeide et al., 2006 for details). Repeatability of IR in previous studies has been high (0.99 in Nordeide et al., 2008, see also Nordeide, 2002, and Nordeide et al., 2006).

The first five minutes of each subtrial, starting when the focal fish was released from the transparent Plexiglas tube (see Fig. 1), were analysed using the VLC Media Player. The monitor was set in black and white mode during this analysis to reduce potential subjectivity from the analyser, since this step hindered distinguishing between both white and green light and between the Normal-grey and the Normal-red dummies. A total of 60 trials were carried out of which 9 were removed from the final data-set due to the sticklebacks being either infected by the endo-parasite Schistocephalus solidus (Eucestoda) (3 individuals), or due to technical problems during recording (2 specimens) or opening of the video-files of sub-trials (4 specimens). This left us with data from all four sub-trials from 51 trials, 29 males with total length 55.2 (S.D. ± 4.84) mm and 22 females with total length 55.6 (± 7.69 mm), respectively. During the video-analyses we quantified seven different parameters in each subtrial. First, we quantified to which of the two dummies in a pair of dummies (Normal-grey (NG) vs. Normal-red (NR) or Normal-grey (NG) vs. Large-red (LR)), each stickleback approached first after being released from the transparent Plexiglas tube (TPT). This preference was defined by which of the zones “A” or “B” containing one of the dummies - the focal fish entered first (see Fig. 1). Second, we quantified the time spent in the neutral zone before the focal fish for the first time swam into one of the two zones (Zone A or B in Fig. 1) containing a dummy. Third, we quantified near which of the two dummies the focal stickleback spent the most time. This was done by noting in which of the zones “A” and “B” the focal fish was located every 15 s during the first five min (a total of 20 observations) after being released to swim freely in the experimental tank. Fourth, we quantified close to which of the dummies the focal specimen preferred to pick its first Chironomidae-larvae. Fifth, we quantified the time spent before this first feeding (in the previous point). Finally, we quantified the Intensity of red (IR) of the sticklebacks’ pelvic spines and the body length of each stickleback.

Most statistics were carried out with IBM SPSS Statistics version 27, whereas Cohen’s d and Cohen’s h and statistical power were estimated using the pwr package in R version 4.0.2 (R Core Team, 2020). Significance level was set to 0.05, and all p-values were two-tailed except the χ2 = tests which are one-tailed (Sokal and Rohlf, 1981). Statistical tests were carried out on females and males separately, and additionally on pooled data from both sexes. Exceptions were tests of “First feeding close to a dummy” (see Results, and Table 1 and Fig. 4) where we only tested after pooling data from both sexes due to the relatively low sample size (n), and test of “Effect of the intensity of red pelvic spines of the focal sticklebacks” (Fig. 5) where the response variable (IR) differed significantly between the sexes and thus data were not pooled. Binomial tests were used for frequency data. T-tests or Mann-Whitney U-tests were used to compare measurement variables (time spent and the intensity of red pelvic spines) relative to the choice between different dummies, after testing data for deviation from normal distribution using the Kolmogorov-Smirnov-test. Box-Whiskers plots show 10th and 25th, median value, and 75th, 90th percentiles. A reviewer suggested that we carry out a linear model for each response variable like : “Behavior ~ sex + dummy treatment + light treatment + spine intensity + body size”. Our main argument why this is probably not a good idea are: (i) There are two pairs of dummies involved in this experiment: NG – NR, and NG – LR. A single multivariate linear model for each response variable would also compare the behaviour of the focal fish towards one dummy in one pair of dummies with the behaviour towards another dummy in the other pair of dummies. This does not make sense, and it would flaw our results. Additional challenges are: (ii) Which of several models to pick for presentation when five predictors and their interaction terms are involved and a 0-model being the best model when using an AIC-approach. (iii) Collinearity is involved, and (iv) we would have to remove the power-analyses.

The null hypotheses were not rejected in most of the statistical tests: t-tests, Mann-Whitney U-tests and binomial tests, in this study (see Results). Thus, we estimated statistical power (the probability 1 – β of correctly rejecting a false null hypothesis) for some of the tests (see “Results” and “Power analysis” below). Power is a function of effect size, and a larger difference between the two means to be compared in t-tests, or more specimens preferring one particular of the two zones with the dummies at the expense of the other zone in the binomial tests, would increase the power of our tests. Thus, we iteratively increased the effect size until the statistical tests turned out as marginally significant (p < 0.05) and then we re-estimated statistical power using this new effect size. (i) When increasing the effect size in the t-tests we kept the sample sizes and standard deviations from actual data sets unchanged. We also kept one of the estimated sample means constant, whereas the other mean was gradually changed until the difference between means became marginally significant (p < 0.05), when tested by t-test. We then estimated Cohen’s d based on this new simulated mean difference and pooled sample standard deviations, in accordance with Cohen (1988). Finally,  statistical power was calculated for two-sided t-tests, using the pwr.t2n.test of the pwr package, by inserting actual sample sizes, the newly calculated Cohens d and a significance level of 0.05. (ii) Similarly, to estimate statistical power of the binomial tests we gradually increased the difference in number of observations between the two groups while keeping the total sample size (n) constant, until the p-value from the binomial test turned out to be slightly significant. Then we used these new counts (which were now significantly different) to estimate Cohen’s h and statistical power.

 By these simulations we were able to estimate the magnitude of effect sizes (difference between two means in t-tests, or counts in binomial tests) to detect significant differences between the groups. Effect sizes of Cohen’s d and Cohen’s h less than 0.5, between 0.5 and 0.8, and above 0.8, are considered as “small, “medium”, and “large”, respectively (Cohen, 1988). Finally, statistical power was estimated using the new simulated parameters.

The study was carried out in accordance with ethical guidelines stated by the Norwegian Ministry of Agriculture and Food through the Animal Welfare Act. According to these guidelines, we were not supposed to - and therefore do not - have a specific approval or approval number.

Read the "Read me first"-file.