Exaggerated and conspicuous sexually selected traits are often costly to produce and maintain. Costly traits are expected to show resource-dependent expression, since limited resources prevent animals from investing maximally in multiple traits simultaneously. However, there may be critical periods during an individual’s life where the expression of traits is altered if resources are limited. Moreover, costly sexual traits may arise from sexual selection acting both before (pre-copulatory) and after mating (post-copulatory). Gaining a robust understanding of resource-dependent trait expression therefore requires an approach that examines both episodes of sexual selection following resource limitation during different times in an individual’s life. Yet few studies have taken such an approach. Here, we examine how resource restriction influences a set of pre- and post-copulatory traits in male pygmy halfbeaks (Dermogenys collettei), which invest in sexual ornaments and routinely engage in male-male contests and sperm competition. Critically, we examined responses in males when resources were restricted during development and after reaching sexual maturity. Both pre- and post-copulatory traits are resource-dependent in male halfbeaks. Body size, beak size, courtship behavior, and testes size were reduced by diet restriction, while, unexpectedly, the restricted-diet group had a larger area of red color on the beak and fins after diet treatment. These patterns were generally consistent when resources were restricted during development and after reaching sexual maturity. The study reinforces the role of resource acquisition in maintaining variation among sexual traits.

Study population and housing conditions

Experimental fish were generated from F1 and F2 descendants of wild-caught halfbeaks collected from Sungai Tebrau, Malaysia, that were bred in an aquarium facility at Stockholm University, Sweden. The adult fish, which live for up to ~3 years in captivity (personal observation), used to produce the experimental subjects were kept in mixed-sex groups in 160 L aquaria containing ~2 cm of gravel, aeration, and live and artificial plants. Fish were maintained on a 12:12 light:dark cycle at ~26˚C. Offspring were collected either from mixed-sex groups or individually housed gravid females (females were separated from offspring shortly after birth to prevent infanticide) and then used in one of the two diet manipulation experiments described below. The two experiments were performed consecutively over the course of 6 months in 2019 and 8 months in 2020, respectively. Experiments were approved by Stockholm Animal Research Ethical Board (permit number 2393-2018 and 3967-2020).

Experiment 1: Adult diet experiment

In this experiment, halfbeaks were reared on different experimental diets treatments after reaching sexual maturity (henceforth called the adult diet experiment). Offspring were reared until sexual maturity under standard laboratory conditions (see above). Following birth, up to five offspring (~3-5 mm in length) were housed together in 4 L tanks for up to two months, during which they are fed exclusively live Artemia salina nauplii (henceforth referred to as artemia) twice per day. 20-30 fry from different 4 L tanks that were born within the same calendar month were then moved to 72 L tanks until the onset of sexual maturity. During this period of development, juveniles were fed ad libitum with a mixture of ground flake food and freeze-dried artemia twice per day, 6 days per week. In addition, previously frozen and thawed Drosophila melanogaster and live artemia were fed to fish once and 3-4 times per week, respectively. At the onset of sexual maturity (~2 months of age), males were identified by the developing of the modified anal fin (the andropodium) used for copulation (Meisner and Burns, 1997). Following identification, males were removed from the 72 L tanks and reared in male-only tanks (55 or 160L, 15-30 males per tank) until ~3–4.5 months of age (ensuring sexual maturity) when they were allocated to an experimental diet treatment.

To experimentally manipulate adult diets, sexually mature males (n=70) were taken from male-only tanks at age ~3–4.5 months and placed in individual 4 L tanks (25 x 16 x 12 cm) containing ~2 cm of gravel, four floating plastic plants, 2-3 snails and constant aeration. A floating circular feeding arena (a hollow plastic tube) was placed in the center of the tank. The feeding arena was used to ensure that fish could easily find and access the food. Fish were allocated to one of two experimental diet treatments: a high or restricted quantity diet. There was no difference in body size between groups prior to diet treatment (linear model; F=0.11, df=1, p=0.74). Of the original sample size of 70 males, 4 died during diet treatment (2 from high diet and 2 from restricted diet treatment), reducing the final sample size to 66 males (33 in each treatment). In the adult high-diet treatment (n=33), each week males were fed six fruit flies (D. melanogaster) per day (3 in the morning, 3 in the afternoon) for five days and 3 flies per day for two days (total per week: 36 flies). In the adult restricted-diet treatment (n=33), each week males were fed one fruit fly per day for 5 days and 2 flies per day for two days (total per week: 9 flies). We used only fruit flies to manipulate adult diets because they present a simple way to control food quantity. Males were maintained on these diet treatments for 28-31 days. Note that assays were initiated around day 21 of the diet treatment and continued until termination at day 28-31. In all tanks, approximately 10% of the water was changed three times per week, using a flow through system to minimize disturbance. The adult diet experiment was performed in 4 blocks (n=16-18 males per block, equalized between treatment groups).

Assays (described below) were performed in the following order, each separated by 1-3 days: female mate choice assay, courtship assay, competition assay, morphological and post-copulatory measurements (lateral pictures for body length, beak length, and coloration traits, body mass measurement, sperm assays, and testes mass measurement).

Experiment 2: Developmental diet experiment

In this experiment, halfbeaks were reared on different experimental diets treatments until reaching sexual maturity (henceforth called the developmental diet experiment). New-born fish (n=111 fry < 3 days old, unsexed, i.e., a mix of males and females) were collected from stock tanks and placed in 4 L experimental tanks. For the first 28 days of the experiment, juveniles were kept in groups of 2-4 individuals and fed an ad libitum diet of live artemia twice per day. This initial ad libitum diet stage was done to minimize juvenile mortality, which is greatest during the first month after birth. After these initial 28 days, juveniles were placed in individual 4L tanks containing 2 cm of gravel, plastic plants, 2-3 snails and constant aeration. Juveniles do not eat D. melanogaster and were therefore maintained on a diet of live artemia. Juveniles were allocated to either a high or restricted-diet treatment. In the developmental high-diet treatment, fish were fed 3 ml of the solution obtained from 6 ml artemia cysts hatched in 1200 ml water solution twice per day for 6 days each week (12 x 3 ml feedings per week). Fish in the restricted-diet treatment were fed increasing amounts of food as they grew following the initial 28 days of ad libitum feeding to account for increases in fish size throughout the experiment. The same concentration of artemia was used throughout the experiment. During the first, second, and third month of the diet treatment, juveniles in the restricted-diet treatment were fed 3 ml of the artemia solution i) once per day, 3 days per week (3 x 3 ml feedings per week), ii) once per day, 5 days per week (5 x 3 ml feedings per week), and iii) twice per day for 1 day per week and once per day for 5 days per week (7 x 3 ml feedings per week), respectively. Fish were maintained on these diet treatments for a total of 3-3.5 months, which included the last 1.5 week during which assays were performed.

Halfbeaks live in shoals and interactions with conspecifics are likely a normal part of their development (e.g., Devigili et al. 2021). Therefore, following isolation in individual rearing tanks (i.e., after the initial 28 days of group rearing), all juveniles were exposed to a conspecific once every three weeks (i.e., at week 3, 6, 9, 12, and 15 of diet treatment). During these exposures, one juvenile fish (the “visitor”) was placed into a transparent plastic cylinder (12 cm in diameter), which was subsequently placed inside the tank of another juvenile fish (the “host”) for 4-6 hours (approximately 10AM-3PM). The cylinder was perforated, with ~2 mm holes placed around the cylinder, to allow visual and olfactory cues to be assessed between the visitor and host fish, while preventing direct physical interactions. All fish were fed at least 1.5 hours before the exposures to minimize the risk of live artemia being left in the home tanks when the visitor was added. Each juvenile acted as the visitor or host an equal number of times over the course of the experiment. During each social exposure, juveniles were paired with a novel individual from the same experimental block (see below).

Of the initial 111 juveniles that entered the experiment, 10 died over the course of the experiment (8 pre-treatment, 1 from high-diet treatment, 1 from low-diet treatment), reducing the sample size to 101 fish (mix of males and females). The experiment was performed in three blocks, where fry born within two weeks were pooled into a single block and assayed together. Each of the blocks contained between 32-36 juveniles, with diet treatments equalized within each block. When focusing only on males, this translated into 18-25 males per block. The final sample size in the developmental diet experiment was 50 males (n=27 in the restricted diet treatment and n=23 in the high diet treatment). At 4-4.5 months of age, the assays were initiated. Note that since all assays take approximately 10 days to complete per block, diet treatments continued during this time. The order of the assays was the same as in the adult diet experiment (see above) with the exception that a lateral picture (see below) was captured the day before female mate choice in order to produce the size-matched pairs of males.

Male body size and external morphology

Body size and external morphology was quantified for males in both experiments using photographs. The lateral side of each male was photographed using a digital camera (Canon 800D with macro lens EF-S 60 mm) inside a photo chamber (30 × 20 × 20 cm) under standard light conditions. Each image included a scale bar. From these images, body length was measured in ImageJ (v1.52i; Schneider et al. 2012) from the anterior part of the eye to the caudal peduncle (Supplemental Fig. 1). This measure of body length excludes the jaw, which is variable among males. We hypothesized that the length of the beak, which may be important in both male-male competition and female mate choice (Berten and Greven 1991; Greven 2006), could exhibit resource dependence and this could be independent of the rest of the body. We therefore measured the beak length separately from the rest of the body, from the anterior tip of the beak to the anterior part of the eye (Supplemental Fig. 1). Additionally, we measured the area of the yellow and red coloration on the caudal, dorsal, and anal fins, black coloration on the dorsal fin, and red and black coloration on the beak using the polygon selections tool in ImageJ (v1.52p) by tracing around the colored areas of the respective body parts. For the analysis, we included only normally developed beaks (i.e., straight), reducing the sample size to 98 males (adult diet experiment: high n=30; restricted n=23; juvenile diet experiment: high n=22, restricted n=23).

To examine coloration on the fins and beak, we used a principal component analysis (PCA) to reduce the dimensionality of the data, thereby reducing the number of statistical tests (and the associated risk of Type-II errors) and account for potential collinearity among variables. The total amount of red, yellow, and black coloration on the fins and red and black coloration on the beak were entered into a PCA. The PCA returned two PCs (henceforth coloration PC1 and PC2) with standard deviations ≥ 1, which together accounted for 56% of the total variance and were used for further analysis (Supplementary Table 1). Loading values of each PC are considered to contribute to that PC when the loading values are 70% of the variable with the highest loading. Coloration PC1 was primarily loaded by yellow fin coloration and black beak coloration, while coloration PC2 was primarily loaded by red beak coloration and the total red coloration on the fins (Supplementary Table 1).

Body mass was measured by placing fish individually in a small dish containing water on a previously tared balance (Mettler Toledo XS105). This procedure was repeated for two independent mass measurements per individual and an average of the two measurements was used for analysis. Body length and mass were measured at the start and end of the diet treatments in the adult diet experiment, while measures were only taken at the end of the diet treatments in the developmental diet experiment. We estimate body condition using Fulton’s condition factor (K = (body mass / length³) * 100), a common proxy of condition in fish (Nash et al. 2006).

Male courtship behaviors

Male courtship behaviors were recorded using a free-swimming assay where one focal male could interact with one female. Individual males from each of the adult and juvenile diet treatment groups were added to an experimental tank (dimensions 40 x 24 x 30 cm) filled to a water depth of 15 cm, containing two plastic plants and ~1 cm of gravel. A sexually mature (i.e. >4 months old) virgin female randomly chosen from a female-only stock tank was then added to the experimental tank. The male and female were separated by a transparent plastic divider and left undisturbed for 1 hour, allowing the fish to habituate to the experimental tanks. After the habituation period, the transparent divider was lifted using a pulley system, allowing the male and female to interact. During a 20 min observation period, we recorded a number of behaviors. As discrete behaviors, we recorded the number of male circling, where the male swims around the female's head in a semicircle, which may be an initial mate assessment behavior, and matings, a rapid movement (~40-80 ms) where the male twists his body to attempt to insert the andropodium into the females genital pore (Greven 2010; note that it is impossible to distinguish between successful or unsuccessful mating attempts solely from behavioral observations). As continuous behaviors, we recorded the time a male was observed swimming under the female while positioned ventrally and posteriorly with his head directly below the female’s genital pore, nipping, where the male is swimming under the female but now moves his upper and lower beaks, rapidly clapping the upper and lower beak together, and checking at genital pore, where the male makes physical contact with the female’s genital pore using his upper jaw. Behaviors were recorded in real-time by one experimenter (E.F.I.). Fish were fed ~2 hours prior to all courtship assays. Trials for all courtship assays were performed and analyzed blind to the treatment group by giving the males a unique identifier code.

We summed the behaviors measured as counts (circling and mating) to generate a total number of sexual behaviors index. Similarly, all behaviors measured as durations (time spent under female, nipping, or checking at the genital pore) were summed together to generate a total duration of sexual behaviors index. After summing the behaviors, these indices were divided by the amount of time that the focal male was aware of the female, which was obtained by subtracting the time from the start of the trial until the male and female first interacted from the total time of the trail. Courtship trials were aborted if the male or female showed signs of stress (such as repeatedly swimming against the glass; n=6) or if either fish acted aggressively towards the other (usually the female towards the male; n=4). Thus, the final sample size for the courtship assays was 106 trials (adult diet experiment, high n=30; restricted n=30; developmental diet experiment, high n=20, restricted n=26).

Male-male competition behaviors

Competition behaviors were video-recorded in a free-swimming assay where one male from the high-diet treatment and one from the restricted-diet treatment were allowed to interact. To examine the effect of resource restriction on competitive interactions removing body size-mediated effects, males in each trial were size matched prior to the assay with a maximum difference of 3 mm in length allowed between the males in the pair. On average, the difference in length within the pair was 0.44 ± 0.08 mm (mean ± SE) in the adult diet experiment and 1.43 ± 0.16 mm in the developmental diet experiment. In two blocks of the developmental diet experiment, there were two more fish in the juvenile restricted-diet treatment than in the juvenile high-diet treatment. Therefore, these four fish created pairs with two males of the same treatment. These pairs were treated the same as regular pairs (i.e., went through the competition assay and were recorded), in order for all fish to have the same experience for subsequent assays. These two recordings were not included in the analysis (see below).

The competition assays were recorded from above with a camera (Point Grey Grasshopper 3 4.1 megapixel camera with Fujinon CF25HA-1 lens) placed 1.5 meters directly above a circular experimental tank (50 cm in diameter, containing 4 cm of water). Each male in a pair was first placed in a separate opaque cylinder (15 cm in diameter) inside the experimental tank and allowed to acclimate for 15 min without physical or visual contact with the other male. After the acclimation period, the recording was initiated, and the cylinders were lifted by hand. The recording proceeded for 20 minutes. The video recordings were subsequently scored blind by one experimenter (E.F.I.). In each trial, a randomly chosen focal individual was selected and their behaviors were recorded by keeping track of the focal fish frame by frame (we refer to the other fish in the pair as the ‘rival fish’). We scored the number of displacements (where one fish swims up to the other from the side or front, causing the other to divert), the number of lunges (one fish charges quickly at the other fish and makes physical contact with its beak somewhere on the body of the other fish), and the duration of swimming behind (the fish is swimming up to five body lengths behind the other, facing in the same direction, measured in seconds). All behaviors were scored from the perspective of the focal fish, recording the number and duration of behavior performed by the focal fish and directed at the rival fish or the number and duration of behaviors that were experienced by the focal fish (i.e., performed by the rival fish). Lunges and displacements were considered aggressive behaviors. In total, 58 trials were recorded (33 in the adult diet experiment and 25 in the developmental diet experiment). Scoring of behaviors stopped before the end of the recording (i.e., reducing the observation time) if the observer lost track of the IDs of the experimental individuals (n=10).

For analysis of competitive behaviors, we converted the behaviors experienced by the focal individual to performed by the rival, in order to have a comparable value for the males of the two treatments. Therefore, the data is analyzed per individual, rather than per recording.  Since each recording is represented two times in the data, we also control for recording ID in the analysis. The behaviors measured as counts (displacements and lunges) were summed per individual. Count behaviors and the duration behavior were divided by the total trial time. Trials had to be a minimum duration of 10 min of interactions to be included in the analysis, thus excluding four shorter trials. Three recordings were excluded due to technical errors while recording. Two recordings were excluded since the males in the pairs were from the same diet treatment (see above), leading to a final sample size of 100 individuals (adult diet experiment, high n=30; restricted n=30; developmental diet experiment, high n=20, restricted n=20; from 50 recordings).

Female preference assay

A dichotomous choice assay was used to assess whether male resource-dependent traits influence the outcome of female mate preferences. For both the adult and developmental diet experiments, one female and two males (one from each diet treatment) were placed in an experimental tank (45 x 25 x 20 cm), consisting of one main chamber (45 x 15 x 20 cm) and 3 adjacent, isolated stimulus chambers (each 15 x 10 x 20 cm, see Fig. 2 in Reuland et al. 2019). The same individuals were used in the pairs for the female mate choice assay as for the competition assay (i.e., one high and one restricted diet male).  A focal virgin female between 4 and 6 months old was randomly chosen from a female-only stock tank and placed in the main chamber, and the two males were randomly placed each in either the left or right stimulus chamber. The middle stimulus chamber contained no fish but was filled with gravel and water for the purpose of spatially separating the right and left chambers. Transparent glass walls separated the main chamber and the three smaller stimulus chambers. Thus, the focal female could only use visual cues to assess the stimuli males. Opaque walls separated the three stimulus chambers to prevent visual contact between stimuli males. The entire experimental tank was surrounded by opaque dividers to reduce the potential for external factors to influence the fish’s behavior.

Before each trial, fish were given 60 min to habituate to the environment, during which time visual contact between all chambers was prevented by an opaque divider. After the habituation time, the divider between the main and stimulus chambers was lifted using a pulley system and trials were recorded for 60 min using a GoPro Hero 5 Black digital camera (GoPro, Inc., San Mateo, CA, USA) or a webcam (Logitech C920 1080P HD, Carl Zeiss Tessar) positioned 30 cm above the tank. The duration of time a female spent in the left or right association zones (an area of 5 x 15 cm in front of each stimulus chamber) was used as a measure of female preference for the male in that chamber. Association time is a commonly used proxy of mating preference, which is predictive of preference during real mating interactions (Bischoff et al. 1985; White et al. 2003; Walling et al. 2010). To quantify informed choice (or preference) we only considered association times after the choosing individual has visited the association zones of both stimulus individuals. The water in the tank was changed between trials to avoid any remaining olfactory cues from the previous trial.

In total, 55 trials were conducted. In seven cases the female never visited the second male, or visited him late into the video, resulting in less than 10 min of informed choice (n=3). Two recordings from the developmental diet experiment where both of the two males in the pair were of the restricted diet treatment were removed (n=2). Additionally, trials were excluded if the female or one or both males behaved in a stressed manner (n=4). The final sample size consisted of 39 trials (adult diet experiment, n = 22; developmental diet experiment, n = 17).

Sperm quality and testes size

At the end of the diet treatments, and following behavioral assays, males were euthanized in a benzocaine solution (150 µl benzocaine per 1 ml ethanol), washed in deionized water and placed on a glass slide containing ~1 ml of a 9% saline (NaCl) solution and viewed under a dissection microscope (S9 stereo microscope, Leica Microsystems, Wetzlar, Germany). Male halfbeaks produce sperm bundles called spermatozeugmata (Grier and Collette 1987). Sperm bundles were extracted into the saline solution by applying gentle pressure with a blunt instrument to the abdomen and moving towards the posterior part of the testicular duct, the channel that transports sperm from the testes to the andropodium, a modified anal fin used to transfer sperm to females (Greven 2010). 20 µl of the sperm/saline solution was then transferred to an Eppendorf tube containing 20 µl of activation solution (Hank’s Balanced Salt Solution (HBSS), Sigma-Aldrich, United Kingdom) in a 1:1 ratio of sperm/saline to HBSS for use in subsequent analyses.

Sperm velocity, viability and morphology were measured from separate subsets from the activated sperm/saline solution using an integrated semen analysis system (ISAS, v. 1.2.33, PROiSER R+D, Paterna, Spain). Since they are measured from different subsamples, sample size may differ between traits (see below). To assess sperm velocity, 3 µl of the sperm/saline solution was transferred to two wells of a 12-well multitest slide (MP Biomedical) coated with a 1% polyvinyl alcohol (PVA) solution and covered in a previously coated PVA cover glass to prevent cells sticking to the glass (Wilson-Leedy and Ingermann 2007). Sperm swimming parameters were characterized using computer-assisted sperm analysis (CASA) software (UB 200i Series Microscope and C13-ON camera, PROiSER R+D, Paterna, Spain). Of the total of 116 males from both experiments, the velocity subsamples of 8 males had insufficient sperm to facilitate sperm measurements, reducing the sample size to 108 males (adult diet manipulation, high n=30; restricted n=32; juvenile diet manipulation, high n=23, restricted n=23). Only velocity samples with a minimum of 20 sperm cells measured were used, and the threshold for defining static cells was predetermined at 25 µm/s for VCL. Applying these exclusion criteria reduced the final sample size to 93 males (adult diet manipulation, high n=26; restricted n=23; juvenile diet manipulation, high n=22, restricted n=22). From these males, sperm velocity was measured for a mean (± SE) of 102.76 ± 12.17 and 183.59 ± 15.52 sperm cells per male in the adult and juvenile diet manipulation experiments, respectively. We focus on three commonly used, colinear sperm velocity metrics obtained from CASA, including the average path velocity (VAP), straight-line velocity (VSL) and curvilinear velocity (VCL). These three sperm motility measures were reduced using a PCA. The PCA returned one PC with standard deviation ≥ 1, which accounted for more than 96% of the total variance and was used for further analysis (Supplementary Table 2). Sperm velocity PC1 was loaded by all three measures of sperm velocity (VSL, VCL and VAP) (Supplementary Table 2).

Sperm viability, quantified as the proportion of live sperm cells in each ejaculate, was measured using a live/dead cell viability kit (VitalTest, NordicCell, Denmark). A 14 µl subsample from the sperm/saline/HBSS solution was transferred to a new Eppendorf tube and 1.6 µl of propidium iodide and 0.5 µl of acridine orange was added. After adding the dyes, 14 µl (two samples of 7 µl each) of stained sperm solution was transferred to a microscopic slide and left for 1-2 minutes in darkness, after which it was placed under a fluorescent microscope and images were captured (×200 magnification; UB 200i Series Microscope and C13-ON camera, PROiSER R+D, Paterna, Spain). Viable sperm with an intact outer membrane are labelled green by membrane-nonpermeable dye (acridine orange) while dead or inviable cells with disrupted membranes are labelled red with membrane-permeable stain (propidium iodide). Sperm viability was then calculated based on a mean (± SE) of 244.6 ± 34.7 cells (range 25 - 1383) in the adult diet-manipulation experiment and 311.3 ± 4.2 cells (range 74 - 1046) in the developmental diet experiment. The total number of cells that was counted in the samples is accounted for in the analysis (see below). Of a total of 116 males from both experiments, the viability subsamples of 10 males had insufficient sperm, reducing the sample size for sperm viability to 106 males (adult diet experiment, high n=30; restricted n=29; developmental diet experiment, high n=23, restricted n=24).

For sperm morphology measurements, images of sperm cells were captured from each male’s ejaculate (×400 magnification; UB 200i Series Microscope and C13-ON camera, PROiSER R+D, Paterna, Spain). Pictures were analyzed in ImageJ to measure length of head, midpiece, and flagellum (mean number of sperm cells analyzed per male in the adult diet experiment = 18.6 ± 0.5 SE; range: 4–20; in the developmental diet experiment = 18.5 ± 0.3 SE; range: 13-20). The three measures of sperm morphology were condensed using a PCA, which returned one PC (henceforth sperm morphology PC1) with standard deviation ≥ 1, which accounted for 42% of the total variance and were used for further analysis (Supplementary Table 3). Sperm morphology PC1 was loaded negatively by flagellum length and positively by head and midpiece length (Supplementary Table 3). Of the total of 116 males from both experiments, the sperm morphology subsamples of 10 males had insufficient sperm, reducing the final sample size for sperm morphology to 107 males (adult diet experiment, high n=30; restricted n=30; developmental diet experiment, high n=23, restricted n=24).

After sperm assays, the male was dissected and the testes were removed under a dissection microscope (Leica S9i; Leica Microsystems Ltd., Heerbrugg, Switzerland) and weighed (Mettler Toledo XS105 scale). We removed one male (adult high-diet treatment) from the analyses as his testes were not properly developed which reduced the sample size to 115 males (adult diet experiment, high n=29; restricted n=30; developmental diet experiment, high n=20, restricted n=26).

Statistical analyses

To account for block effects, that can emerge from sequential sampling, and to allow for statistical comparisons between Experiment 1 and 2, all traits, apart from the response variables in the female mate choice analysis, were converted to standardized trait values prior to analyses by dividing each trait value by the mean of the block for that trait. Since we were explicitly interested in the effects of both diet manipulation treatments (i.e., high vs. restricted diet) and when the diet experiments were applied (i.e., adult vs. juvenile), all models initially included diet treatment, experiment, and their interaction as fixed factors (unless otherwise stated).

First, we validated that male body condition was influenced by the diet treatment. To do this, we use a linear model with male condition (Fulton’s K) as the response variable and the diet treatment, the experiment (adult or developmental diet), and their interaction were entered as fixed factors. In the adult diet experiment, data on male condition was also available at the start of the diet treatment. Therefore, we compared adult condition (response variable) before and after the diet treatment in a linear mixed model with diet treatment, timepoint (at the beginning or end of the diet treatment), and their interaction, as fixed factors, and male ID as random factor.

Next, we examined if diet treatment and timing of diet manipulation (adult or developmental diet) influenced the expression of putative pre-copulatory sexually selected traits (beak length and coloration area) as well as male courtship (total count and duration of sexual behavior indices) and competitive behaviors (total count and duration of competitive behaviors). These (apart from competition behavior) were analyzed using linear models that included the trait as a response variable and diet treatment, experiment, and their interaction as fixed factors. The interaction was removed if non-significant. Competition behavior was analyzed using a linear mixed model to additionally include recording ID as random factor since the analysis is done per individual and not per recording. Body length was added as a covariate to models examining beak length to account for allometric effects. In models with coloration (coloration PC1 and PC2), body length and beak length were added as covariates. In models assessing male courtship behaviors, the total number of courtship behaviors were log10-transformed and the total duration of courtship behaviors was square-root transformed to improve the fit to a normal distribution.

Next, we examined whether females preferentially associate with the male of one treatment over the other using a strength of preference (SOP) score for the left male (arbitrarily defined without regard to diet treatment) as the time a female spends in the association zone of the left male divided by the total amount of time the female spend in the left and right association zones. SOP values range from 0 (females only spend time in the association zone of the male on the right) to 1 (females only spend time in the association zone of the male on the left). We then fit a linear model with SOP (left male) as the response variable, with diet treatment and experiment as fixed factors. To examine if male morphological traits influenced female preferences, we first calculate the difference in morphological trait values between the left and right male regarding body length, beak length, red coloration on the fins, red on beak, black on beak, yellow coloration on the fins, and black coloration on the fins. A linear model was fitted with SOP for the left male as the response variable and the differences in trait values together with experiment as fixed factors. The results did not change when we fitted each trait in separate models.

Finally, we examined the effect of diet restriction on sperm velocity, sperm viability, sperm morphology, and testes mass. For sperm viability, we calculated the proportion of live sperm in the samples from each male. Post-copulatory traits were assessed using linear models that included the trait as response variable and diet treatment, experiment, and their interaction, as fixed factors. Body length was added as a covariate to models examining testes mass to account for allometric effects. The model with sperm viability was weighted by the total number of cells counted (i.e., sum of live and dead cells per sample).

All analyses were performed in R 4.0.2 (R Core Team 2020) using the lm function from the stats package for linear models. The lmer and glmer functions from lme4 package (Bates et al. 2015) were used for the mixed models.  The Anova function from the car package was used to obtain significance effects (Fox and Weisberg 2011). Model fit was assessed using model diagnostics and residual plots from base R. The emmeans package (Lenth 2021) was used to perform post-hoc Tukey tests to identify differences in main effects using. In all cases, non-significant interaction terms were removed from models and simplified models are presented.

See README file.