Exposure to multiple environmental stressors is a common occurrence that can affect organisms in predictable or unpredictable ways. Hypoxia and turbidity in aquatic environments are two stressors that can affect reproductive behaviors by altering energy availability and the visual environment, respectively. Here we examine the relative effects of population and the rearing environment (oxygen concentration and turbidity) on reproductive behaviors. We reared cichlid fish (the Egyptian mouthbrooder, Pseudocrenilabrus multicolor) from two populations (a swamp and river) until sexual maturity, in a full factorial design (hypoxic/normoxic x clear/turbid) and then quantified male competitive and courtship behaviors and female preference under their respective rearing conditions. Overall, we found that the rearing environment was more important than population for determining behavior, indicating there were few heritable differences in reproductive behavior between the two populations. Unexpectedly, males in the hypoxic rearing treatment performed more competitive and courtship behaviors. Under turbid conditions males performed fewer competitive and courtship behaviors. We predicted that females would prefer males from their own population. However, under the hypoxic and turbid combination females from both populations preferred males from the other population. Our results suggest that reproductive behaviors are affected by interactions between male traits, female preference, and environmental conditions.

Rearing Experiment

Fish were collected from one swamp site (Lwamunda) and one river site (Ndyabusole) in the Lake Nabugabo region of Uganda during May 2018. The swamp site is characterized by hypoxic (0.79 ± 0.1 mg/L, mean ± SE point in time measurements) and clear (2.02 ± 0.3 NTU) conditions, while the river site is normoxic (6.84 ± 0.2 mg/L) and moderately turbid (18.81 ± 1.1 NTU; mean ± SE); measurements were taken between June and August in 2015, 2016, and 2017 from Oldham (2018). The wild-caught fish were transported to the laboratory for a full-sibling split-brood rearing experiment. The two populations were held separately and allowed to breed until they produced ten independent broods (10 different male-female pairs) from each population. To ensure that each brood was produced by different parents, one male and three females from a single population were isolated until the first brood was observed. Pseudocrenilabrus multicolor are maternal mouthbrooders. After a female was determined to be holding a brood (distended jaw, refusal of food, and anti-social behavior) she was isolated to continue brooding. When the female released her brood (approximately 2-3 weeks), the young were housed in separate aquaria for one week then the brood was split randomly in a full-factorial design with two factors (oxygen treatment and turbidity treatment) with each having two levels (hypoxic or normoxic and clear or turbid, respectively). Broods from each treatment (oxygen and turbidity) and population (swamp or river) were spread evenly across 80, 19-liter aquaria to account for minor differences in temperature and light availability (10 tanks per population/treatment). Weekly water changes were conducted in each tank to maintain water quality.

In the hypoxic tanks, oxygen was gradually reduced over a period of one week by bubbling nitrogen gas into the water and placing bubble wrap over the surface of the water. To maintain normoxic (mean ± SE: 7.52 ± 0.001 mg/L O₂), and hypoxic conditions (mean ± SE: 2.27 ± 0.01 mg/L O₂), DO concentrations were measured once a week in normoxic tanks and three to seven times a week in hypoxic tanks using a YSI Pro2030 multimeter probe (see Supplementary 1 for full summary of rearing conditions). DO was adjusted as needed by bubbling ambient air or nitrogen gas into the tanks. In the turbid tanks, we gradually increased turbidity over a period of one week by adding ~1.5 mL of bentonite clay solution (70 g in 700 mL water) each day. To maintain clear (mean ± SE: 1.31 ± 0.03 NTU) and turbid (mean ± SE: 16.6 ± 0.14 NTU) conditions, turbidity was measured once a week in clear tanks and one to two times a week in turbid conditions using a Hach 2100Q portable turbidimeter (Supplementary 1). Turbidity was then adjusted as needed during weekly water changes, or for turbid tanks, by adding bentonite clay solution. Temperature was consistent across all tanks (mean ± SE: 24.9 ± 0.02°C).

The number of fish in each brood ranged from 27-105 (mean ± SE: 42.25 ± 4.65), so the initial number of fish in each tank ranged from 6-25 fish. Fish from each brood were randomly divided between treatments. Note that at one week old we could not determine sex, so sex ratios varied across tanks. To decrease and homogenize density, the number of fish in each tank was reduced to ten at six weeks and further reduced to six fish at 20 weeks post-release. The fish were reared under their respective treatment conditions until they were ~1.5 years old when they were used for the behavioral experiments outlined below. At the time of the behavioral experiments, all fish were sexually mature, e.g., males had red spots on anal fin, blue lips, and yellow coloration on ventral surface.

Male-Male Competition

To test the effects of oxygen, turbidity, and population on aggression, we conducted male-male competition trials. Male competition trials were conducted in the same environmental conditions that the fish were raised under (Summary of trial rearing conditions; Supplementary 2). Hypoxic conditions were achieved by bubbling nitrogen gas into the trial tanks and placing bubble wrap over each compartment to help maintain low oxygen conditions, while normoxic tanks had air stones to maintain their respective treatment conditions. Bentonite clay solution was added to turbid trial tanks to increase turbidity. Males were size-matched (less than 8% difference by standard length; distance from snout to caudal peduncle, range 4.4-6.3 cm; mean ± SD: 5.13 ± 0.46 cm) within trials and were from different broods of the same population so that siblings were not used in a trial together. DO, turbidity, and temperature measurements were taken from the center of the tank at the start of acclimation and immediately after the trial (for conceptual diagram see Figure 1A; for snapshot from trials see Supplementary Figure 1). An opaque barrier was placed between the two males (barrier likely did not block olfactory cues) during an overnight acclimation (18-22 hours). Note that because sex ratios varied between tanks, our pre-experimentation housing was not standardized. Accordingly, social status of males and prior experience of males and females likely differed across trials. To initiate the trial the opaque barrier was replaced with a clear one so that the males could interact without direct contact to prevent injury, and trials were filmed from the side for 30 min. Full water changes were performed between trials. Trials were only included in the analysis if the males approached each other (n = 33 trials analyzed out of 60; 2-5 trials per treatment, sample sizes in Supplementary Table 2). Note, that while not quantified formally, there was generally higher mortality in the hypoxia turbid combination, particularly in the river population. Therefore, sample sizes in this treatment combination (river/hypoxic/turbid) in the male-male competition trial were low (two trials) due to a lack of size-matched males that could be tested. We quantified the number of lateral displays (where the male places himself perpendicular to the opposing male), charges (rapid movement towards the opposing male with mouth closed), and bites (rapid movement towards the opposing male with mouth open), as well as the time active (i.e., moving in the water column). All videos for mate choice and male competition were analyzed by one observer using BORIS (Friard and Gamba 2016). The observer was blind to the population of fish in the trials.

Mate Choice

To test the effects of oxygen, turbidity, and population on reproductive behaviors in males and females, we compared female preference and male courtship behaviors between fish reared under the four different treatment combinations and from the swamp and river populations. Specifically, we conducted mate choice trials on fish under the treatment conditions that the fish was reared under (e.g., a female raised in hypoxic/clear conditions was placed in a mate choice tank with the choice of two males, one from each population – swamp and river, that were also raised under hypoxic/clear conditions). The male from the same population as the female is referred to as the native male while a male from the different population is referred to as non-native male. All fish within a trial were from different broods so that siblings were not used in a trial together. Immediately before the overnight acclimation (18-22 hours), the fish were photographed to quantify color (see below) and males were size-matched (less than 8% difference by standard length; range 3.9-6.1 cm; mean ± SD: 4.98 ± 0.49 cm).

In each trial, the trial tank was divided into three compartments (Figure 1B). Sand was included in the trial tanks so that males could build territorial pits overnight. We randomly assigned two males from the same rearing treatment but different populations (swamp population and river population) on either end of the trial tank. Males were isolated from each other and not able to physically interact. The female was placed in the center compartment, also physically isolated from both males. Only females that were not mouthbrooding were selected for use in trials. However, we did not record whether females were gravid at the time of the trial, so female receptivity to male courtship likely varied. A piece of PVC pipe in the middle of the tank was provided as a shelter so the female could choose not to interact with either male. The fish were allowed to acclimate to the tank overnight during which time opaque barriers prevented the fish from seeing each other (barriers likely did not prevent olfactory cues). After acclimation, the opaque barriers were removed (leaving clear barriers), and the fish were filmed from the side for 30 min to record all movement and interactions. Some males were reused, but they were given a least one week in between testing, and males never encountered another male more than twice.  Only trials where females left the hiding spot were included in analysis (n = 81 trials analyzed out of 105; 10-11 females per treatment/population Supplementary Table 2).

The treatment conditions were created and maintained as in competition trials (described above). Videos were recorded from the side and analyzed using BORIS by one observer that was blind to the population. In each videotaped trial, we recorded the amount of time a female spent within 7 cm (~1.5 body lengths) of each male. For males, we quantified the number of lateral displays and quivers (shaking movement of the whole body), as well as time spent active.

Male Color

Male color was calculated using a previously established protocol (Maan et al. 2004). Briefly, before overnight acclimation, males were placed in a small, clear box (i.e., photo cuvette) filled with water and gently pressed against the front of the box using a grey board to be photographed (Canon PowerShot SX740 set to automatic). Using Adobe Photoshop, the photos were white balance corrected using a white standard placed on the front of the cuvette. The area of the fish (excluding fins and eyes) containing red (hue = 0-26 and 232-255, saturation = 40-97%) and yellow (hue = 27-45, saturation = 40-97%) pixels relative to the total number of pixels of the fish body was quantified. This value was used to calculate the overall % red + % yellow color to evaluate overall differences in red and yellow coloration as these colors typically resulting from carotenoid pigments may be correlated with fitness proxies and female preference in cichlids (Maan et al. 2006; Sefc et al. 2014).

Statistics

All analyses were conducted in R, version 4.3.0 (R Core Team 2019), and we used the lme4 package, version 1.10-35.1 (Bates et al. 2015) for generalized linear mixed models (GLMM) and linear mixed models (LMM). To determine the potential independent and interacting influences of oxygen and turbidity on behavior, we modeled these treatments as separate fixed factors and included an interaction term between oxygen and turbidity in all models. We also included population and standard length in all models. Brood was included as a random effect in all GLMM’s and LMM’s to account for relatedness between siblings. All data and analyses are available on Dryad.

For the male competition trials, we modeled the number of competitive behaviors (# of lateral displays + charges + bites) and activity (time spent moving through the water column). All models included oxygen, turbidity, the interaction between oxygen and turbidity, and population as fixed factors, standard length as a covariate and trial and brood as random effects. Due to overdispersion, a GLMM with a negative binomial distribution was used to model the # of competitive behaviors while activity (s) was modeled using an LMM.

Color data (% red + % yellow coloration) for males used in the mate choice trials was analyzed using a LMM which included oxygen, turbidity, the interaction between oxygen and turbidity, and population as fixed factors, standard length as a covariate, and brood as random effect. Because some fish were used in more than one trial (90 unique males were used in mate choice trials), we only included the color measurement for the male’s first trial in this analysis.

For males in the mate choice trials, we also tested for an effect of oxygen, turbidity, population, and standard length on the number of male courtship behaviors (# of lateral displays + # of quivers) and activity (time spent moving through the water column). Because some males were used in more than one mate choice trial (90 unique males were used in mate choice trials; 8-13 males per population/treatment combination, Supplementary 3), we only analyzed male behavior from each male’s first mate choice trial. All models included oxygen, turbidity, the interaction between oxygen and turbidity, and population as fixed factors and standard length as a covariate. A GLMM with a Poisson distribution and a log link was used to model the # of courtship behaviors. Activity was modeled using an LMM.

We quantified female interest in male courtship as the time a female spent within 7 cm (approximately 1.5 body lengths) of either male using a LM with oxygen, turbidity, the interaction between oxygen and turbidity, and population as fixed factors. We also modeled the preference of a female for males from her own population using a LM. Preference was quantified by subtracting the time spent with the non-native male from the time spent with the male from her own population. A positive value indicated that a female spent more time with male from her own population. We calculated a PCA loading factor that combined standard length and the overall % of red + yellow color, as color and size are positively correlated in males of this species. We also calculated relative courtship behavior (# of courtship behaviors by male from native population- # of courtship behaviors by non-native male). A positive value for relative courtship indicated that the male from the local population engaged in more courtship behaviors. The relative differences of the PCA loading factor and time spent in courtship were included in the model of female preference.

We used a SEM to understand the direct and indirect effects of oxygen, turbidity, and population on female preference. We used the results from our GLMM and LMM’s on male traits as well as previous studies to develop our a priori model (Figure 2; (Atkinson and Gray 2022; Gray et al. 2012; McNeil et al. 2016)). Our SEM was modeled using lavaan (Rosseel 2012). To test the assumption of multivariate normality, we examined the χ² Q-Q plot from the MVN package (Korkmaz et al. 2014). To improve assumptions of multivariate normality, female preference (time spent with native or non-native male) and male courtship behavior (# of courtship behaviors) were square root transformed. Transformed data met assumptions of multivariate normality. For variables that were expected to be related, but without an obvious causal relationship, we included the covariance between residuals (male courtship, male size, and the time a female spent with either male). Whether our a priori model adequately fit the data was determined by examining the Model χ²  (P > 0.05), and approximate fit indices (Comparative Fit Index > 0.95, Standardized root mean square residual < 0.10) (Grace 2020). We did not remove nonsignificant paths from our final model.

Excel is required to open the files dfcomp, dfcourtbyfish, and dfcourtbytrial.

R is required to run the scripts Competition_script.R, Courtship_script.R and Rearing_env_data_script.R