Study animals
Mosquitofish is a poeciliid fish species native to North America, and it has been intentionally introduced in many countries with the aim of controlling mosquitos [42]. Poeciliid fish are characterized by promiscuity, internal fertilization, ovoviviparity, and sexual dimorphism, with males being smaller than females [43]. At sexual maturity, female mosquitofish possess two gravid spots on the posterior of their abdomens, while males have a gonopodium modified from the anal fin. The mating system in this species is non-resource-based and promiscuous, in which males do not court, but instead, sneakily approach females from behind and attempt coerced copulations [42]. After internal fertilization, the fertilized ova hatch within the female ovary often during a period of 22–25 days [42]. The brood size of a female depends on its body size, with larger ones giving birth to more new-borns, which are approximately 6–8 mm in length [42]. The time for fry to reach sexual maturity varies from one month to several months, depending on the water temperature. Standard body length (SL) at sexual maturity is usually larger than 15 mm in males and 17 mm in females [42].
A total of 1500 new-borns generated from 150 wild-caught females were uniformly reared in 30 net tanks (80 × 80 × 80 cm, mesh size: 0.177 mm) in an outdoor artificial pond on the campus of Anhui University from May to July 2018 [44]. They were fed brine shrimp nauplii until two weeks old and thereafter fine-grained commercial food (TIDDLER, Weifang YEE Pet Products Co., Ltd., China; 42% crude protein, 5% crude fat, 5% crude fiber, and 11% ash). The water temperature during the rearing period ranged from 20°C to 32°C, and the pH ranged from 7.4 to 7.6. Apart from the additional food, the rearing conditions were the same as those of their conspecifics already living outside the tanks in the pond for several years, avoiding behavioral abnormalities that might arise in laboratory conditions [45].
In July, 80 sexually mature males (SL > 15 mm), indicated by a clear apical hook at the gonopodium tip [42], were randomly net caught from the rearing tanks and were randomly categorized into two groups with each containing 40 individuals. One group was tested in the following Experiment 1 (Expt 1) and the other in Experiment 2 (Expt 2, Fig. 1). To avoid any external stimulus, the males were individually kept in black, opaque, cylindrical tanks (height: 9 cm; diameter: 15 cm; hereafter, holding tank) with a black, opaque, cylindrical refuge chamber (height: 5 cm; diameter: 7 cm) placed in the center. The holding tanks were filled with oxygenated tap water, and the fish were given > 24 h to acclimate to the chamber before the experiments. Then each group of the 40 males were evenly divided into 10 subgroups that were tested separately with different healthy, active, virgin females as stimuli. Specifically, for each subgroup of males used in Expt 1, one pair of mature females with different body lengths (22 mm vs. 18 mm) were used as stimuli, and for each subgroup in Expt 2, one pair of mature females with the same body size (22 mm) were used. Because the intensity and size of gravid spots are linked with developmental stages and clutch size of female live-bearing fish and thus may influence mate choice of males [46, 47], we paired females with similar sizes and intensities of gravid spots. Because the handlings of fish during experiments were always in water, and the fish were always given enough time to acclimate, no stressful responses were observed in the stimulus females.
To mimic two distinct degrees of female activity for males to choose from, females were restricted (inactive female, IF) or not (active female, AF) by a transparent plastic cylinder (6.5 cm diameter × 8 cm high) as required during the experiments. After a period of acclimation, the IFs adapted to the trap and lowered their activity levels with no obvious abnormal behaviors. Instead of using actual activity levels that may vary significantly and not as expected during experiments, this man-made neat difference in activity levels between stimulus females can generate more robust results. Because the cylinders used to trap the females were transparent and were separated by plexiglass, males could not perceive that the IFs were constrained.
General experimental procedure
To test the effect of the personality of male mosquitofish on their mate choice, open arena assays were firstly used to measure shyness and activity twice for each male in two successive days (see the subsection of personality measurements for details). Subsequently, mate choice experiments were carried out for each of the two groups of 40 males, respectively (see the subsection of mate choice experiments for details). The first mate choice experiment (Expt 1, Fig. 1a) tested males’ preferences for body sizes (22 mm vs. 18 mm) of mature females and whether these preferences were influenced by female activity levels (active vs. inactive) and male personality traits. The second one (Expt 2, Fig. 1b) tested males’ preferences for females with the same body size but different activity levels (active vs. inactive) and whether their preferences changed when the inactive female increased its activity level.
The experiments were carried out in a white opaque plastic tank (37 cm long × 30 cm wide × 20 cm high, Fig. 2) in the same laboratory under enough light and constant temperature (26°C). A camera (Sony HDR-CX510, 55× extended zoom, Sony Corporation, Tokyo, Japan) was fixed above the tank to record the behaviors of the subjects throughout. To avoid potential disturbances, the experimenters were shielded from the experimental apparatus by a 1.5-m high opaque curtain during each trial. To minimize observer bias, blinded methods were used when all behavioral data were recorded and/or analyzed. At the end of all experiments, each male was gently placed against the transparent wall of a glass tank to measure its SL (accurate to 0.1 mm), and all the fish were released to the rearing pond.
Personality measurements
In the assays for shyness and activity, a black, opaque, cylindrical refuge chamber (the same refuge placed in the holding tank, hereafter, starting refuge) was fixed at the far end of the experimental tank (Fig. 2a). A sliding trapdoor (3 × 3 cm) on the side of the starting refuge facing the arena was connected to a piece of fishing line that allowed experimenters to remotely open the refuge to allow fish to emerge from the chamber and move toward the arena without disturbance. The tank was filled with oxygenated tap water to a depth of 3 cm, and the water was exchanged after each subject was tested. Males were not fed beyond 12 hours before the personality assays.
At the beginning of the assay, a male was gently transferred from the holding tank to the closed starting refuge in the arena. The subject was allowed to acclimate for 5 min, and then the trapdoor of the refuge was remotely opened and was not closed until the end of the trial. Following previous studies [48-50], shyness was measured as the time taken by the subject to swim out of the refuge, that is, for its whole body to cross the trapdoor. All the subjects emerged from the refuge within 6 min. After the male left the refuge, it was video-tracked continuously for 10 min to record its movement trajectory. At the end of the trial, the subject was immediately transferred to its holding tank. A total of 600 image stacks were extracted from the 10-min movement videos (one frame per second); Image J (http://rsbweb.nih.gov/ij/) was used to delineate each subject’s movement pathway, of which the total length was used to quantify its activity level.
Mate choice experiments
To carry out the mate choice trials in a dichotomous chamber, the white opaque plastic tank was divided into three compartments separated by plexiglass, only allowing visual contact between fish in different compartments. At the center of each compartment, there was a transparent plastic cylinder (6.5 cm diameter × 8 cm high) connected to a piece of fishing line, by which the experimenters could remotely pull up the cylinder to allow the trapped fish to swim freely. Two dark lines were drawn on the bottom of the tank to demarcate a neutral zone (11 cm wide) and two preference zones (5-cm wide; Fig. 2b). The tank was filled with 7-cm-depth oxygenated tap water, which was changed after each trial. One hour before the mate choice experiments, males were fed to avoid the effect of hunger.
Expt 1: In this experiment, four successive sub-experiments were designed to study whether male’s personality and female’s activity level influence male preference for female body size. Firstly, the two females (active large female, AL_F; active small female, AS_F) were separately placed outside the cylinders in the end compartments of our experimental setting, and a male was trapped inside the cylinder in the central compartment. After a 10-min acclimatization period, the male was remotely released to allow it to choose between the two females, and its movement behavior was video-tracked for 12 min (Expt 1a). Secondly, the male and the smaller female (18 mm, IS_F) were trapped inside their cylinders, while the larger female (22 mm, AL_F) was outside the cylinder. After 10 min for acclimatization, the male was released to swim freely and was video-tracked for 12 min (Expt 1b). Thirdly, the male and the 22-mm female (IL_F) were trapped, while the 18-mm female (AS_F) swam freely. After 10 min, the male was released and video-tracked for 12 min (Expt 1c). Fourthly, the male and the two females (IL_F, IS_F) were all trapped initially, and after 10 min, the male was released and video-tracked for 12 min (Expt 1d). To make the same experimental conditions for all males, they underwent the same order of the above treatments. The non-randomized order may have mask effect on the treatments (i.e. the order effect). However, the 10-min acclimatization between treatments may reduce the possible order effect.
Expt 2: Two sub-experiments were used to explore whether males (another 40 individuals) choose mating partners according to male’s personality traits and female’s activity level. Firstly, two stimulus females of the same body size (22 mm) were separately placed in the end compartments, with the IF restricted inside its cylinder while the AF freely swam outside. A male was restricted inside the cylinder in the central compartment for 10 min to acclimate and observe the activities of both females. Then, the experimenter allowed the male to swim around freely and video-tracked its movement for 12 min (Expt 2a). Secondly, the male was restricted inside the cylinder again, and the IF was released to swim freely (its label changed to ‘IFtoAF’ accordingly). After 10 min for acclimatization, the male was released and video-tracked for 12 min (Expt 2b). To avoid potential side-biases, the locations of AF and IF were exchanged between trials.
Statistical analyses
Following the protocol of Dingemanse and Dochtermann [51], a bivariate linear mixed-effects model was fitted using the R package MCMCglmm for the males used in Expt 1 and 2, respectively, to measure the repeatability of each behavior (i.e. personality) and the among-individual correlation between the two behaviors (i.e. behavioral syndrome). The two loge-transformed behavioral traits were concurrently included as the response variables in each model with individual ID as a random effect. The bivariate models were run for 220,000 iterations after 20,000 burn-in iterations and were thinned by 25 iterations. Model convergence was confirmed by the Gelmin-Rubin diagnostic test (function gelman.diag).
Male mate preference for certain females was determined as the association time, that is, the time a male spent within the preference zone near each female stimulus. The association time is widely used to measure mating preferences in female and male live-bearing fish [17, 52-56]. To further avoid disturbances from experimental operations, the middle 10 min of each 12-min free-choice period was used for the analyses. Time 1 (T1) was defined as the association time of the male with the large (or active) female and T2 with the small (or inactive) female. Male’s strength of preference (SOP) for large versus small (in Expt 1a and 1d) or active versus inactive females (in Expt 2a) was calculated as: SOP = (T1 − T2) / (T1 + T2). The SOPs ranged from −1 to 1, with larger values indicating a stronger preference for large (or active) females. The SOPs were tested using the Shapiro-Wilk test and were found to be normally distributed (all p values > 0.25).
The mean values of shyness and activity measured for each male in the two personality trials were used in the following analyses. A t-test was used to compare the differences in body length, shyness, and activity between the males used in Expt 1 and 2. Principal component analysis (PCA) was performed for the two groups of males, separately, to summaries the two highly correlated behavioral traits into two independent principal components (PC1 and PC2). The repeated-measures ANOVA in R package nlme and post hoc Tukey’s test with a Bonferroni correction were used to test and compare the differences in time spent by males in mate choice assays among the three different zones (i.e. neutral zone and the two preference zones). A paired t-test was used to investigate the change in males’ association time with the IF in Expt 2a after it was released in Expt 2b (IFtoAF).
General linear mixed models were performed by R package lme4 to fit the effects of male body length, PC1, and PC2 on male mate preferences for the larger (Expt 1a and 1d) or the active female (Expt 2a). The SOPs were used as the dependent variables, male body length, PC1, and PC2 as the fixed effects, and the pair ID of stimulus females as the random effect in the models. Two-way interactions between body length and PC1 and PC2 were initially included but were excluded in the final models because there were no significant effects. There were no significant correlations between body length and PC1 (Pearson correlations, Expt 1: r = 0.23, p = 0.151; Expt 2: r = 0.12, p = 0.462) and PC2 (Expt 1: r = 0.09, p = 0.579; Expt 2: r = -0.05, p = 0.754). Thus, there was no problem with multicollinearity associated with the models. All analyses were carried out using R 3.6.1 [57], and data are displayed as mean ± standard error.