Strength of sexual selection and sex roles vary between social groups in a coral reef cardinalfish
Data files
Apr 08, 2024 version files 344.31 KB
-
Aggression_Ruegeretal_2023.xlsx
-
Bateman_code-2.R
-
Bateman_Ruegeretal_2023.xlsx
-
Bateman_standardized.csv
-
Cardinalfish_sex_roles_aggr_dimorph_skew_updated.R
-
demographics.csv
-
Group_aggression_2024.csv
-
Parentage_Ruegeretal_2023.csv
-
Parentage_Ruegeretal_2023.xlsx
-
parents.csv
-
README.md
-
Trait_Ruegeretal_2023.xlsx
Abstract
The strength and direction of sexual selection can vary among populations. However, spatial variability is rarely explored at the level of the social group. Here we investigate sexual selection and sex roles in the paternally mouthbrooding, socially monogamous, and site-attached pajama cardinalfish, Sphaeramia nematoptera. Females were larger, more aggressive, and had a longer dorsal fin filament, indicating reversed sex roles. At the scale of social groups, we show the Bateman gradient and reproductive variance depending on the sex ratio and size of the groups. In small and medium-sized groups with balanced or male-biased sex ratios, Bateman gradients were steeper for females, whereas gradients were equally steep for both sexes in large groups or when the sex ratio was female-biased. For both sexes, reproductive variance increased with group size and with a higher male-to-female sex ratio. In S. nematoptera, mating opportunities outside the socially monogamous pair appear to impact sexual selection. We conclude that the strength and direction of sexual selection can be masked by social dynamics in group-living species when considering only population and large-scale demographic processes.
README: Strength of sexual selection and sex roles vary between social groups in a coral reef cardinalfish
Rueger et al. (Forthcoming 2024). Strength of sexual selection and sex roles vary between social groups in a coral reef cardinalfish. The American Naturalist.
The dataset contains the data underlying the Bateman gradient, reproductive output, and reproductive skew analyses (Bateman gradient file), as well as the data underlying the behaviour and trait analyses (aggression and Trait data files). We have also included the code for the Bateman gradient analysis and figures.
Data were collected over five periods, spanning two years; October - November 2012
(observational period: 32 days), February - March 2013 (21 days), July - August 2013 (16 days), March - April 2014 (23 days), and September 2014 (9 days), in Kimbe Bay, Papua New Guinea (5°30’S, 150°05’E).
Description of the data and file structure
Each dataset is available as a separate Excel workbook and includes the data in tab 1 and variable descriptions and measurement units are provided in tab 2.
Population structure data ("dmographics.csv")
Variables include:
Year - the year data was recorded
Group - social group ID
ASR - adult sex ratio (number of males: number of females)
adult - number of adults in the group
Aggression recorded between individuals ("Aggression_Ruegeretal_2024.xlsx")
Variables include:
Date - sampling time period
Site - social group ID
Tag - individual fish ID
SL - standard length (mm)
Sex - sex of the individual
Status - paired/unpaired status as determined by behavioural observations
Aggr - total number of aggressions
AggrOther - aggression towards heterospecific
AggrHomo - aggression towards individuals of the same sex
AggrHetero - aggression towards individuals of the opposite sex
Clutch - number of clutches
NoM - number of mates
ASRcat - Adult sex ratio category
adultcat - category for the number of adults in the group
density - density of adults (number of adults/group territory size)
ASR - Adult sex ratio
adult - number of adults in the group
Aggression recorded per group, summarized ("Group_aggression_2024.csv")
Variables include:
Date - study period
Group - social group ID
ASR - adult sex ratio
adult - number of adults in the group
group.agr - number of aggresive behaviours counted for each group in each study period
Trait data (including standard length, weight and dorsal fin filament) ("Trait_Ruegeretal_2024.xlsx")
Variables include:
DF - dorsal fin filament length (mm)
weight - wet weight (g)
sex - sex as determined by dissection and observing gonads
maturity - determined by standard length (mm)
Reef - inshore patch reef ID in Kimbe Bay, Papua New Guinea
Site - group ID
ID - individual fish ID
SL - standard length (mm)
TL - total length (mm)
Genopytes for the parentage analysis, which is what reproductive output is based on ("Parentage_Ruegeretal_2024.xlsx")
Variables include:
OffspringID - ID of embryo analysed
clutch - clutch ID
Date - Date sampled
Year - Year sampled
Reef - reef ID in Kimbe Bay, Papua New Guinea
Group - group ID
Female - female identified as parent
Male - male identified as the parent
COLONY parentage assignments to calculate reproductive skew ("parents.csv")
Variables include:
Offspring ID - unique identifier for each offspring genotype
Clutch - clutch ID
Date - sampling period
Year - year of sampling period
Reef - the name of the reef where the clutch was sampled
Group - group ID of the S. nematoptera group where the clutch was sampled
Female/ Male - assigned parents according to COLONY
Bateman gradient data (including the number of offspring and mates for each individual and the number of adults and sex ratio in each group) ("Bateman_Ruegeretal_2024.xlsx")
Variables include:
ID - individual fish id
Year - year of sampling
sex - sex of the individual
Group - social group ID that the individual is located in
offspring - number of offspring recorded for the individual during the sampling year
mates - number of mates recorded for individuals during the sampling year
ASR - adult sex ratio in social group
adults - number of adults in a social group
ASRcat - adult sex category
adultcat - number of adults category
Standardized Bateman data ("Bateman_Ruegeretal_2024.xlsx" with meta-data and the working csv file "Bateman_standardized.csv")
Variables include:
ID - individual fish id
Year - year of sampling
sex - sex of the individual
Group - social group ID that the individual is located in
offspring - number of offspring recorded for an individual during the sampling year
mates - number of mates recorded for individuals during the sampling year
ASR - adult sex ratio in social group
adults - number of adults in social group
ASRcat - adult sex category
adultcat - number of adults category
R code for Bateman gradient analyses ("Bateman_code-2.R")
Bateman_code has the following sections, with annotation, description, and session info included:
- Library
- Data import and useful functions
- Bayesian modeling
- Model investigation 4a. Investigate model quality 4b. Investigate model output
- Plotting
R code for all other analyses ("Cardinalfish sex roles_aggr_dimorph_skew.Rmd")
Contains the following sections, with annotation included:
- Methods
- Group size and structure 2a. Model 2b. Figure
- Sexual dimorphism 3a. Model 3b. Figures
- Aggression 4a. Model 4b. Figure
- Reproductive Skew 5a. Models
Methods
Study system and tagging
This study was conducted over five periods, spanning two years; October - November 2012 (observational period: 32 days), February - March 2013 (21 days), July - August 2013 (16 days), March - April 2014 (23 days) and September 2014 (9 days), in Kimbe Bay, Papua New Guinea (5°30’S, 150°05’E). To reflect approximately the same observational timespan and thus enable comparisons of reproductive output, data were grouped by year. Social groups of the pajama cardinalfish, Sphaeramia nematoptera, were defined as occupying the same distinct patch of the stony coral Porites cylindrica at a minimum distance of 2m from another group (Fig. 1a). A total of 379 individuals from 18 groups at 5-17 m depth on five reefs were included in the study (see map in Rueger et al. 2019). All fish were caught using handnets and diluted clove oil solution (Munday and Wilson, 1997), and individually marked using Visible Implant Elastomer (VIE) tags (Northwest Marine Technology). For each individual, we recorded a unique identifier, their life stage, and sex (see Group size and structure), behaviour (see Behavioural observations), collected fin clips (see Measuring mating and reproductive success), and measured standard length. Other morphological measures (weight and dorsal fin filament) were only collected from a smaller sample (see Sexual dimorphism). This study was conducted in accordance with the James Cook University Ethics Committee, approval number A1847.
Group size and structure
The number of individual fish in 18 spatially different groups were counted and adult and juvenile status were determined based on standard length (Rueger et al. 2016b). Sex was determined by observing the distended buccal cavity during brooding (male) and bulging abdomen shortly before brooding (female) as well as pairing behaviour (Rueger et al. 2016b, 2018). A total of 168 individuals in 18 groups were identified as adults based on standard length with the remaining 211 being juveniles or subadults. Detailed demographics, including adult sex ratio and adult group size (number of adults per group), were gathered for 18 groups. For 10 groups three years of observations were available, for four groups two years of observations were available and for four groups only one year of observations was available. Reproduction was observed for 121 adult individuals for 16 of the 18 groups; in one group reproduction was observed in all three years, in nine groups reproduction was observed in two years and six groups reproduction was observed in only one year. All statistical analyses in this study were carried out in R version 4.0.3 (R Core team 2020). To determine whether the adult sex ratio and group size varied significantly between groups, we conducted a linear mixed model (LMM) analysis using the lme4 package (Bates et al. 2015). We used adult sex ratio and group size as response variables in two separate models, group ID as fixed factor and year as random intercept. Significance tests for LMMs were performed by likelihood ratio tests of the full model with the effect in question tested against the model without the effect. No obvious deviations from homoscedasticity were detected by visually inspecting the residual plots. No outliers or high variance inflation factors (VIF) were detected in any of the best-fit models, using performance (Luedecke et al. 2020). Conditional and marginal R2 were calculated using Nakagawa’s R2 in performance.
Sexual dimorphism
A sample of 47 adult S. nematoptera were caught from multiple groups and euthanized to extract their gonads and determine sex (see Rueger et al. 2018 for detailed methods). These samples were not part of the long-term monitoring effort described above for 18 groups but instead originated from separate groups. There were 20 females and 27 males in the sample. Their standard length (SL; fish length from the tip of the snout to the posterior end of the last vertebra excluding the length of the caudal (tail) fin) and the length of the dorsal filament (Fig. 1d) were measured to the nearest millimetre using calipers and wet weight (mass) was measured using a digital scale to the nearest 0.1g.
We conducted a directed comparison between three morphological traits, SL, mass, and dorsal fin filament, that are: 1) highly distinguishable in this species (dorsal fin filament length) (Fig. 1d), and 2) commonly sexually dimorphic among other species (body size). While dorsal fin filament length may simply be represented by a univariate measurement (mm), body size, in contrast, is better represented as a multivariate morphological trait (Freeman & Jackson 1990). Therefore, we first constructed a multivariate metric of body size by loading SL (mm) and mass (g) into a Principal Component Analysis (PCA) and extracting the first component from the PCA (the allometric size variable) for each individual (henceforth defined as ‘Body PC1’; explaining 97% of the variance in SL and mass measurements). To aid in visual interpretation, PC1 values were normalized between 0 and 100. PCA was conducted in R using the function ‘prcomp’, and mass was not log-transformed before loading into the PCA because both mass and dorsal filament scaled linearly (Linear Model: β = 2.96, t = 10.10, df = 45, p < 0.001, adjusted R2 = 0.687). Sexual dimorphism in body size was tested using a linear model with body size (Body PC1) as the Gaussian distributed response variable and sex as the sole fixed effect predictor.
Sexual dimorphism in dorsal filament was tested using a similar linear model, however, body size was replaced with dorsal filament (mm) as the response variable (Gaussian distributed). To account for the influence of body size on dorsal filament alone, and to test for differences in allometric scaling between body size and dorsal filament between sexes, we included sex (binomial), body size (numeric), and an interaction between sex and body size as fixed effect predictors.
The effects of either ASR or group size were not tested on measures of sexual dimorphism because these measures were not known for all groups the sampled fish originated from. A large part of the sample came from the same group (24 out of 47) which had an ASR of 1.27 and was collected as a whole (all 24 adults). The size distribution for the 24 individuals in this group was similar to the other 23 fish included in this sample (see suppl. Figure S1). The other 23 fish in the sample were collected opportunistically from multiple groups of unknown ASR.
Behavioural observations
During each of the five observational periods, all individuals from 18 groups were located every two to three days via visual census on SCUBA. On each day we noted whether males were brooding, which individuals were in pairs (see Rueger et al. 2018 for details) and we recorded the subject and object of any agonistic behaviours (chases or bites) over 15-20 min. We compared the number of aggressions recorded per individual between males and females using Fisher’s exact test. To determine whether adult sex ratio or group size influenced the number of observed aggressions, we fitted a generalized linear mixed model in lme4 (Bates et al. 2015) with the number of aggressions per group as response variable, number of adults and ASR as predictor variables and group ID as random effect term. We used a Poisson error distribution and fitted an observation level random effect term to account for overdispersion. To account for one potential environmental factor that could influence mating dynamics, we tested whether using coral size (the size of the coral head occupied by each distinct group) as a covariate improved the model. Including coral size did not improve the model and we, therefore, did not include it in the final model (log-likelihood test of model with coral size vs without coral size; χ2 = 0.368, p = 0.544).
Measuring mating and reproductive success: Genetic parentage analysis
Relative mating and reproductive success were measured by analysing the parentage of embryos and matching them to known adult individuals in the population. A DNA sample was taken from each individual (N = 379) in the 18 groups via a commonly used non-lethal protocol (Dietrich & Cunjak 2006; Beldade et al. 2016; O’Donnell et al. 2017), by clipping their caudal fin using surgical scissors (Fig. 1d). The resulting fin-clips were stored in high grade 95% ethanol. The procedure was carried out during the same sampling effort (approximately 90 minutes) for all group members at the beginning of the first study period. Caudal fin tissue grew back to the pre-procedure size within 2-3 weeks and no adverse effects on survival or behaviour were recorded (TR pers. obs.). Within the study periods, all broods were collected by catching brooding males. Clutches were then subsampled for genetic analysis and individual embryos were stored in high-grade 95% ethanol. To consider the possibility of multiple mothers and fathers in each clutch, approx. 10 eggs were sampled from different parts of each egg mass, including several points on the surface and the centre of the congealed egg mass (see details in Rueger et al. 2019). A total of 1056 embryos from 105 broods carried by 64 males were assayed.
All individuals were genotyped at 23 microsatellite loci with a range of 3 to 34 alleles observed per locus. Four markers that showed high genotyping error (≥ 6%) were excluded from the analyses. The remaining 19 loci had an average genotyping error of 2.2% ± 0.4 SE. Parentage assignments were conducted with the software COLONY v2.0 (Jones & Wang 2010) to identify the most likely mother or mothers, and father or fathers, of the sampled eggs carried by male cardinalfish. We identified the number of genetic mates (mating success) and the number of offspring (reproductive success) for each reproductive individual. Detailed methods of genotyping and parentage analysis as well as marker-specific statistics can be found in Rueger et al. (2015) and Rueger et al. (2019).
Bateman gradient
We investigated the Bateman gradient by estimating the linear regression slope between mating success (number of genetic mates) and reproductive success (number of offspring assigned through parentage) (Lande & Arnold 1983; Arnold & Wade 1984). We compared the slopes between males and females collating all data for each individual for the five observational periods and using Bayesian regression models (brms package, Bürkner 2017). To explore the influence of group size and structure we calculated the mean number of adults for each group (Fig. 1b), and the mean adult sex ratio for each group (Fig. 1c) over all observational periods. We determined the relationship between mating success and reproductive success using group size (number of adults per group) and group adult sex ratio as continuous covariates, and sex as a categorical covariate. We included the four-way interaction between ‘number of mates’, ‘sex’, ‘adult sex ratio’ and ‘number of adults’, the three-way interactions between ‘number of mates’, ‘sex’ and ‘adult sex ratio’, as well as between ‘number of mates’, ‘sex’ and ‘number of adults’, and the two-way interaction between ‘number of mates’ and ‘sex’. To facilitate comparison with different species and group structures, we used standardized Bateman gradients, where we used the relative number of offspring and z-standardised number of mates (Collet et al. 2014; Schielzeth 2010). We also used z-standardised measures of adult sex ratio and adult group size. To account for non-independence of samples we also included group ID as a grouping term. We used weakly informative priors. We ran the model for 10,000 iterations in each of four chains, with a 2,000-iteration warm-up. We confirmed model suitability using standard Bayesian diagnostic tests (Leave-one-out (LOO) Pareto k: 97% <0.5, 3% 0.5–0.7; Rhat: all = 1; Neff ratio: all > 0.9; non-divergent trace plots; acf = low autocorrelation; density plots = unimodal; diagnostic plots in suppl. Fig. S2-3). LOO cross-validation results indicated predictions in the fitted model were robust to pointwise out-of-sample replacement of data points. Because the response variable (number of offspring) consists of count data we also used expected log predicted density (elpd) scores to compare the fit of the Gaussian linear regression model typically used in Bateman’s gradient assessments to one with a Poisson distribution. The more typically used linear regression model fit the data better (elpd difference = -60.7) and facilitates easier interpretation of the Bateman gradient, we therefore chose this model for further exploration. As above, we tested whether using coral size as a covariate improved the models. Eldp scores were lower when including coral size and we did therefore not include it in the final models (Δ ELPD=-2.1, SE=2.9). Using the covariate model constructed from our observations, we compared Bateman gradient slopes (emmeans package, Luedecke 2018) for males and females across nine hypothetical breeding groups with crossed levels of group size (low = 3, medium = 8, high = 15) and adult sex ratios (male-dominated = 1.9, balanced = 1, female-dominated = 0.6). This ‘counterfactual’ modelling is useful for identifying the causal implications of manipulating one or more predictor variables, in lieu of experimental population manipulations, therefore showing the implied predictions for imaginary experiments (McElreath 2020). Model diagnostics are presented in suppl. Figure S 2, 3.
Reproductive variance
We quantified reproductive variance (sums of squares of differences from means in reproductive success - as the number of embryos assigned to each parent) across the full study period for each sex, overall, and within the 16 groups in which reproduction was observed. We used a Bayesian framework, with a Gaussian distribution, to model differences in within-group reproductive variance for each of the sexes, in relation to the interacting effects of standardised group size and adult sex ratio. We standardised within-group measures of group size and adult sex ratio by subtracting the overall mean and dividing the result by the overall standard deviation, of each variable respectively (sensu Jones 2009). The interaction between standardised group size and the adult sex ratio was removed from the reduced model, after showing nonsignificance. A comparison of LOO elpd scores showed this reduced model fit the data better than the full interaction model (diff >2). As above, including coral size reduced model fit and we did therefore not include it in the final models (Δ ELPD=-1.7, SE=1.4). Models were checked for fit as above. Estimated marginal means and credibility intervals were extracted using emmeans (Lenth 2020) and plotted with ggeffects (Luedecke 2018). Model diagnostic plots are shown in suppl. Fig. S 4.
Reproductive skew
To calculate skew in reproductive success (RS), we used the multinomial index implemented in the SkewCalc package (Ross et al. 2020). The multinomial index (M) is related to Nonacs’ binomial index B (Nonacs & Hager 2011) and accounts for heterogeneity in the number of observational periods in which an individual was observed (Ross et al. 2020). We calculated M for males, females, and the population as a whole, as well as for the different adult sex ratios and group size categories for each year. To explore the influence of group size and sex ratio on M, for each of the 16 groups for each of the three years where reproduction was observed (number of groups reproducing n2012 = 6; n2013 = 11; n2014 = 13), we assigned a categorical variable representing group size (small: n2012 = 2, n2013 = 5, n2014 = 4; medium: n2012 = 2, n2013 = 3, n2014 = 8; large: n2012 = 2, n2013 = 3, n2014 = 1), and a category for adult sex ratio (female-biased (<1: n2012 = 1, n2013 = 4, n2014 = 3), equal (1: n2012 = 1, n2013 = 3, n2014 = 6), male-biased (>1: n2012 = 4, n2013 = 4, n2014 = 4)).