SEX RATIO AND DENSITY AFFECT SEXUAL SELECTION IN A SEX-ROLE REVERSED FISH

Authors


Abstract

Understanding how demographic processes influence mating systems is important to decode ecological influences on sexual selection in nature. We manipulated sex ratio and density in experimental populations of the sex-role reversed pipefish Syngnathus typhle. We quantified sexual selection using the Bateman gradient (math formula), the opportunity for selection (I), and sexual selection (Is), and the maximum standardized sexual selection differential (math formula). We also measured selection on body length using standardized selection differentials (s′) and mating differentials (m′), and tested whether the observed I and Is differ from values obtained by simulating random mating. We found that I, Is, and math formula, but not math formula, were higher for females under female than male bias and the opposite for males, but density did not affect these measures. However, higher density decreased sexual selection (m′ but not s′) on female length, but selection on body length was not affected by sex ratio. Finally, Is but not I was higher than expected from random mating, and only for females under female bias. This study demonstrates that both sex ratio and density affect sexual selection and that disentangling interrelated demographic processes is essential to a more complete understanding of mating behavior and the evolution of mating systems.

Understanding the dynamics of animal mating systems is vital to our knowledge of the sexual selection process. It is now recognized that sexual selection and mating systems can vary within and among populations and species (e.g., Emlen and Oring 1977; Gwynne 1984; Rowe et al. 1994; Neff et al. 2008; Mobley and Jones 2009; Olsson et al. 2011). Sexual selection and mating systems can be influenced by a variety of intrinsic and extrinsic factors including parental investment (Trivers 1972; Kokko and Jennions 2008), the spatial and temporal distribution of partners (Emlen and Oring 1977; Shuster and Wade 2003), and food availability (Gwynne 1990). In general, ecological and demographic factors that influence the environmental potential for polygamy will have an impact on the direction and strength of sexual selection (e.g., Emlen and Oring 1977; Mobley and Jones 2009; Olsson et al. 2011).

One demographic factor known to influence sexual selection is the operational sex ratio (OSR), which is equivalent to the ratio of receptive females to receptive males (Emlen and Oring 1977). Sexual selection theory predicts that the sex in excess will compete more intensely for access to mates as the OSR becomes increasingly biased (Emlen and Oring 1977; Kvarnemo and Ahnesjö 1996). Several studies have found support for this prediction; males compete more in male-biased OSRs (Gwynne and Simmons 1990; Enders 1993; Kvarnemo et al. 1995; Jirotkul 1999a; Grant and Foam 2002; Weir et al. 2011), and may also increase their courtship effort (Carrillo et al. 2012; de Jong et al. 2012). At highly biased OSRs, however, aggression may decrease or level off as territory or mate defense becomes too costly (Mills and Reynolds 2003; Clark and Grant 2010). In addition, sex roles can be reversed (competitive females and choosy males) when the OSR is female-biased (Gwynne and Simmons 1990; Forsgren et al. 2004; Clark and Grant 2010). Studies of the effect of OSR on the distribution of mating or reproductive success show that measures of sexual selection such as the Bateman gradient (βss, the relationship between mating success and reproductive success), the opportunity for selection (I, the standardized variance in reproductive success), and the opportunity for sexual selection (Is, the standardized variance in mating success) will increase for males as the OSR becomes more male biased and the opposite for females (Jones et al. 2004, 2005; Klemme et al. 2007; Croshaw 2010). However, this pattern is not always consistent. In water striders, the variance in male-mating success decreases at male-biased OSRs due to a change in female-mating behavior as harassment rates from males increase. As a result, females will accept more matings at male-biased OSRs due to high costs of harassment (Arnqvist 1992; Krupa and Sih 1993). See also Kokko et al. (2012) and references therein.

Effects of population density on sexual selection is also gaining interest in studies of sexual selection (Emlen and Oring 1977; Kokko and Rankin 2006; Mobley and Jones 2007; Head et al. 2008). Theoretical work suggests that the variance in reproductive and mating success should increase with density due to higher competition for partners among the most competetive sex. Some individuals may be able to monopolize more matings at high densities compared to low densities (Kokko and Rankin 2006). However, at high population densities it might be impossible for a few individuals to monopolize matings due to the cost of increased competition (McLain 1992; Rowe et al. 1994; Mills and Reynolds 2003; Knell and Pomfret 2008). Some studies have found an increase in sexual selection in the most competitive sex from low- to high-density populations (Zeh 1987; Soucy and Travis 2003; Tomkins and Brown 2004; Mobley and Jones 2007). However, other studies have found that higher densities reduce sexual selection (McLain 1992; Rowe et al. 1994; Jirotkul 1999b; Knell and Pomfret 2008). Finally, some studies have found no effect of density on sexual selection (Head et al. 2008; Wacker et al. 2013). Hence, the few existing empirical studies show that sexual selection can increase, decrease, or remain unchanged in relation to increased population density depending on the species and specific ecological conditions.

In species with conventional sex roles, Bateman noted that males had increased variance in mating and reproductive success compared to females, and presented this phenomenon as an explanation for why sexual selection is usually stronger in males (Bateman 1948; Arnold and Duvall 1994). In the decades since Bateman's discovery, three important mathematical principles (Bateman's principles) with respect to sexual selection have emerged: (1) the variance in reproductive success (I) is higher in the sex experiencing greater sexual selection, (2) the variance in mating success (Is) is higher in the sex experiencing greater sexual selection, and (3) the correlation between mating success and reproductive success (βss) is stronger in the sex experiencing greater sexual selection (Bateman 1948; Arnold and Duvall 1994; Jones 2009).

The use of Bateman's principles as measures of sexual selection has been criticized (Klug et al. 2010; Jennions et al. 2012, but see also Krakauer et al. 2011). The criticisms mainly concern the issue that I and Is measures the potential for but not necessarily the actual strength of sexual selection (Crow 1958; Wade 1979; Arnold and Wade 1984; Jones 2009), and because a large variance in mating or reproductive success does not necessarily imply a deviation from nonrandom mating (Sutherland 1985; Klug et al. 2010; Jennions et al. 2012). In our study we try to overcome these weaknesses by including I and Is under simulated random mating, and by including measures of selection on a phenotypic trait used in mate choice.

The purpose of this study is to use the sex-role reversed broad-nosed pipefish, S. typhle, to investigate how sex ratio and density influences the strength and direction of sexual selection. The broad-nosed pipefish is well suited for studying this topic. In nature, both the OSR and the population density of breeding pipefish change over the course of the breeding season (Vincent et al. 1994; K. B. Mobley and A. G. Jones, unpubl. data), The OSR is primarily female biased due to male pregnancy but has also been observed to be male biased in the beginning of the breeding season (Vincent et al. 1994).

The broad-nosed pipefish (S. typhle) has exclusive paternal care. Males carry the embryos in a brood pouch and provide nutrients and oxygen to the developing young (Berglund et al. 1986b, 1986a). Both females and males mate multiply, so the mating system can best be described as polygynandrous (Berglund et al. 1988; Jones et al. 1999b). Females can produce twice the amount of eggs a male can carry during one brooding cycle (Berglund et al. 1989). Therefore, the OSR will be primarily biased toward females (Berglund et al. 1986a; Vincent et al. 1994). Consequently, S. typhle has reversed sex roles and males are generally more choosy than females (Berglund and Rosenqvist 1993). However, females as well as males will prefer a larger mate suggesting mutual mate choice for larger mates (Berglund et al. 1986a). Mate choice decisions in S. typhle are influenced by the OSR (Berglund 1994) and the encounter rate with partners (Berglund 1995). An earlier study of the genetic mating system of S. typhle found that manipulating the OSR changes the strength of sexual selection in the predicted direction (Jones et al. 2005). Females compete intensively over males and can interrupt mating attempts by chasing away other females (Vincent et al. 1994; Berglund and Rosenqvist 2001). Females form dominance hierarchies where larger females are dominant and subsequently suppress the reproduction of smaller females (Berglund et al. 1989).

Our experiment consists of breeding populations at two different adult sex ratios (ASRs) and two different densities in a factorial design. We have four predictions in relation to how sex ratio and density will affect the different measures of sexual selection: (1) The relationship between mating success and reproductive success (βss) should be stronger for females in female-biased than male-biased treatments and the opposite should be true in males. Within each treatment we expect the sex in excess to have a steeper βss than the limiting sex. (2) The opportunity for selection (I) and the opportunity for sexual selection (Is) should be higher for females in female-biased treatments than male-biased treatments, and vice versa for males. We also expect that the sex in excess should have higher values of these measures than the limiting sex in each treatment, and that our observed values for I and Is should be higher than the values obtained under random mating simulations in all treatments. (3) Our measures of selection on body length, the maximum standardized selection differential (math formula), selection (s′), and mating differentials (m′) should be higher for females in female-biased treatments than in male-biased treatments and vice versa for males. (4) Our density treatment could either increase or decrease the opportunity to monopolize matings. However, behavioral data from the same experimental set up as in this study suggests that female–female competition is particularly strong in the female-biased high-density treatment (Aronsen et al. 2013). We therefore expect our high-density treatment to increase measures of sexual selection because it might be possible for large and dominant individuals to better monopolize partners at high encounter rates with potential partners. This effect should be more pronounced for females especially under female bias because females are generally more competitive than males (Berglund et al. 2005). Alternatively, sexual selection might decrease at higher densities, perhaps due to a breakdown of dominance hierarchies when encounter rates with competitors are high.

Methods

EXPERIMENTAL DESIGN

We conducted the experiment in the period late April to mid-June in the years 2008–2010 at the Sven Loven Centre for Marine Research, Fiskebäckskil, Sweden. We collected the fish in the Gullmarsfjord on the west coast of Sweden (58°15′N, 11°28′E) by trawling shallow eelgrass meadows (2–5 m depth) before the breeding period started. The fish were kept in 225 L barrels (males and females separated) containing artificial eelgrass and provided with a continuous flow of seawater until the fish were ready to breed (i.e., when males have developed fleshy brood pouches and females have mature eggs as shown by rounded abdomens). We fed the fish three times every day with live and frozen brine shrimp (Artemia sp.) and frozen mysid shrimp (Mysidae). The salinity and temperature followed the natural fluctuations in the study area (the temperature ranged from 10°C to 16°C during the replicates). Light was provided by lamps in the 18:6 light/dark cycle characteristic for the season and study area.

To investigate the effect of density and sex ratio on the genetic mating system of S. typhle, we established breeding populations at two different ASRs and two different densities. Hence our experiment consisted of four treatments: Female-biased/low-density (F-L), male-biased/low-density (M-L), female-biased/high-density (F-H), and male-biased/high-density (M-H). The female-biased ASR consisted of 20 females and 10 males, and the male-biased ASR consisted of 10 females and 20 males. Because all the pipefish were mature adults, the ASR will be directly related to the OSR before mating. The density was manipulated by using tanks of different sizes (differing water volumes and the size of bottom areas) instead of manipulating the number of individuals. This method is preferable because the latter often leads to confounding effects of sex ratio and density. Manipulating the space the individuals can occupy influences the encounter rate between the fish but does not alter the total number of interacting individuals.

At the start of a replicate each fish was measured (to the nearest mm) for standard length (SL) using a measuring board. We choose fish that reflected a natural range of body length among breeding pipefish (Table 2). There were no differences in female or male length, or the variance in length between treatments (P > 0.40 for all tests). There were differences in body length for both males and females between the five replicates (females P = 0.03, males P = 0.001), but no differences between replicates in the variance in length (P > 0.26 for all tests).

We used three white, fibreglass tanks for the breeding populations, each measuring 202 cm (L) × 113 cm (W). The depth of the water column was 60 cm. In the low-density treatments, the breeding populations occupied the entire tank (∼2 m2). We divided one tank in two equal parts and sealed them to prevent water exchange. This ensured that high-density treatments occupied half the area of the tank of the low-density treatment (∼1 m2). Each enclosure was supplied with a constant flow of fresh seawater and a separate drain. We adjusted the flow of water to be roughly equal in all treatments (doubled water flow in low density to compensate for the larger volume of water). The tanks contained 50 (low density) or 25 (high density) evenly distributed tufts of artificial eelgrass to provide shelter for the pipefish. We have limited information about natural densities of S. typhle, but both these densities might be high compared to the wild. One study investigating the density fish species in eelgrass meadows from the same area as in this study indicated an S. typhle density of 129 individuals per 500 m2 (0.26/m2; Phil et al. 2006). However, densities of S. typhle on a smaller scale will likely be higher due to a “patchy” distribution (G. Rosenqvist, pers. obs.).

Each evening we visually estimated how full of eggs each male's brood pouch was to the nearest 10%. Because water temperatures followed seasonal fluctuations and because the temperature influences the activity of S. typhle (Ahnesjö 1995), we did not standardize the time each replicate lasted. Instead, we ended each replicate when mating activity in all treatments had stopped (very few or no eggs deposited during the last 24 h of the replicate).

The experiment was repeated seven times. We conducted three replicates in both of the 2008 and 2009 seasons and one replicate in 2010. However, in the last replicate of 2008 and 2009, many males died before the embryos were sufficiently developed for DNA samples, so the data presented in this study are from the five remaining replicates (20 populations in total). We have one replicate that lasted for 3 days, two lasting 7 days, and two lasting 10 days.

GENOTYPING

After the experiment all females were fin clipped (taking 2–3 mm2 of the caudal fin) and released back into the wild. Pregnant males were kept in barrels according to replicate and treatment and supplied with continuously renewed seawater and artificial seagrass until the embryos had developed visible eyespots (approximately 2 weeks after the termination of a replicate). Males were then euthanized in seawater containing an overdose of the sedative 2-phenoxyethanol (2 mL/L), after which their spinal column was severed above their operculum. Female fin clips and pregnant males were preserved in 95% EtOH.

The male pouch was dissected and the embryos taken out and numbered according to their position in the pouch. Because the embryos are clumped by maternity in the male pouch, genotyping every fourth embryo is sufficient to find all the mothers contributing to the clutch (Jones et al. 1999a; Jones et al. 2005). All parents and offspring were genotyped using three fluorescently labeled, highly variable microsatellite markers: Typh04, Typh16, and Typh18 (Jones et al. 1999a). Extraction of DNA from the embryos/fin clips was accomplished with a 5% Chelex/Proteinase K method (Walsh et al. 1991) or a Qiagen DNeasy® kit. Polymerase chain reaction (PCR) was used to amplify microsatellite loci with Qiagen multiplex PCR kit. We used 10 μL reaction volumes per sample containing 5 μL multiplex solution (containing HotStarTaq DNA polymerase, dNTPs and 3 mM MgCl2), 0.0625 μL of each primer, and 3 μL of genomic DNA. Thermocycling profile for PCR reactions was as follows: denaturation at 94°C for 15:00 min, 25 cycles of 94°C for 0:30 min, 58°C for 1:30 min, 72°C 1:00 min, and a final extension phase of 60°C for 30:00 min. The microsatellite fragment sizes were determined with an ABI 3730 DNA Analyzer (Applied Biosystems, Foster City, CA). Fragment lengths were scored using ABI GeneMapper™ version 4.0 software.

PARENTAGE ASSIGNMENT

We genotyped all 300 candidate mothers, the 264 pregnant males, and 3876 offspring. The mean ±SE number of genotyped embryos per pregnant male was 14.84 ±1.95 (range: 1–42). Parentage was assigned to the embryos using Cervus version 3.0 (Marshall et al. 1998; Kalinowski et al. 2007). The parameters used for the simulations were: 10,000 cycles, one parent known, 100% of candidate parents sampled, 1% genotyping error. Cervus assigned 90% of the offspring to a candidate mother with 95% confidence. For the remaining 10% Cervus assigned a mother but did not give a confidence level. This was usually due to a mismatch on one marker (Typh16), which could be hard to score if the DNA was highly degraded. In these instances, individuals were assigned to a mother using the remaining two markers.

For some embryos, the DNA was not of sufficient quality to obtain a genotype, we then genotyped the neighboring embryo. If that embryo also failed to yield sufficient DNA (N = 38) we allocated this embryo (= four offspring) to the mother assigned to the offspring located before and after in the pouch. In three cases, there was a different mother before and after this embryo and we therefore assigned two offspring to each of these mothers (note that the removal of these embryos from the analysis does not change our results). Some males died before the embryos were sufficiently developed for a DNA sample. In total, we lost data from one male in each of two replicates in the F-L treatment, one male in one replicate in the F-H treatment, one male in two replicates of M-L treatment, and one male in one replicate and two males in one replicate in the M-H treatment. These individuals were excluded from all analyses.

MEASUREMENTS OF SEXUAL SELECTION

We calculated all measures of sexual selection: the Bateman gradient βss, the opportunity for selection I, the opportunity for sexual selection Is, the maximum standardized sexual selection differential math formula, the standardized selection differential on body length s′ and the standardized mating differential on body length m′ as described in Table 1. We calculated math formula (rather than βss based on unstandardized data) using relative values (each individual's value was divided by the population mean) for mating success and reproductive success (Arnold and Duvall 1994). We used linear regressions of relative mating success on relative reproductive success for each sex in each population to find the math formula, which was also used to calculate math formula. Following the technique developed by Lande and Arnold (1983), selection on body length (s′, m′) was calculated by standardizing body length to have a mean of 0 and a standard deviation of unity after log transformation. Individuals without any offspring were included in the analysis, because these individuals are an important part of the variance in fitness (Wade and Shuster 2004).

Table 1. Measures of sexual selection used in this study
MeasurementExplanationDefinitionReference
  1. rs = reproductive success, ms = mating success, math formula and math formula = the mean reproductive and mating success, math formula and math formula = the variance in reproductive and mating success, respectively.

Bateman gradient (βss)The relationship between rs and msThe slope of a weighted linear regression of rs on ms(Bateman 1948; Arnold and Duvall 1994)
Opportunity for selection (I)The standardized variance in rsmath formula(Crow 1958; Wade 1979)
Opportunity for sexual selection (Is)The standardized variance in msmath formula(Wade 1979; Wade and Arnold 1980)
Maximum standardized selection differential (math formula)Maximum strength of precopulatory phenotypic evolutionβssmath formula(Jones 2009)
Standardized selection differential (s′)Directional selection on the phenotypic trait with respect to rsCovariance (standardized trait, rs)(Lande and Arnold 1983)
Standardized mating differential (m′)Directional selection on the phenotypic trait with respect to msCovariance (standardized trait, ms)(Lande and Arnold 1983; Jones 2009)

SIMULATING I AND Is UNDER RANDOM MATING

To simulate expected values of I and Is under random mating, we used an individually based model. In this simulation model, one male and one female were drawn at random and the female transferred a random number of eggs to the male. The fecundity of males and females and the number of eggs transferred during each mating was drawn from a distribution based on the actual numbers observed in this experiment (see Supporting Information). The mating of males and females by randomly drawing one individual of each sex was repeated until either there were no more eggs left in any of the females, or until there was no more space left in any of the males. The number of mates for each individual was recorded in addition to number of eggs transferred/received. The simulation was repeated 10,000 times for each sex ratio treatment, and I and Is were calculated as defined in Table 1.

STATISTICAL ANALYSIS

All statistical analyses were done in R version 2.13.1 (R Development Core Team 2011). For all models that included interactions, we tested whether the fit of the model decreased significantly if the interaction was removed using chi-squared tests. If the P-value of this test was below 0.1, then the interaction was kept. We first investigated differences between treatments (sex ratio and density) in all measures of sexual selection for each sex separately and then tested for differences between the sexes in each of the four treatments separately.

When investigating the effect of sex ratio and density on math formula, we first tested if the relationship between relative reproductive success and relative mating success was positive in all treatments for both sexes using linear regressions. We initially included replicate as a fixed factor in this analysis but the effect of replicate was not significant and was therefore removed from the analysis. We then used the individual slope estimate for each sex in each treatment and replicate when investigating the effect of sex ratio and density on math formula.

We investigated the effects of sex ratio, density, and the interaction between them on math formula, I, Is, math formula, s′, and m′ with a two-way analysis of variance (ANOVA) with sex ratio and density as fixed factors for each sex separately. We then tested for differences between the sexes within each treatment individually with a one-way ANOVA. To improve the distribution of the residuals, I and Is were square root transformed and the math formula was squared. We checked the assumption of homogeneity of variances between treatments before the analysis by means of a Levene's test for equality of variances.

To test whether the s′ and m′ were significantly different from 0, we obtained 95% confidence intervals (CIs) of both measures calculated across replicates by resampling the data 1000 times by bootstrapping. Parameter estimates were then regarded as statistically significant if 0 was not included in the CI defined as the 0.025 and 0.975 quantiles from the sample of the posterior distribution.

Because the variances of the simulated values of I and Is were very low (range SE: 0.0008–0.002), we tested for differences between the simulated values of I and Is and the observed values in each treatment and sex with the observed data tested against the mean of the simulated data by the means of a one sample t-test.

Results

The mean OSR ([the number of males that still have empty pouch space]/[males with empty pouch space + all females]) and body length are reported in Table 2. The mean ±SE of all measures of sexual selection are reported in Table 3. Some males in the male-biased treatments remained unmated (M-L = 16, M-H = 17), or did not fill up their pouches (mean fill % for mated males, M-L = 76.7 (±3.1), M-L = 72.6 (±3.2). In the female-biased treatment, a few males remained unmated (F-L = 3, F-H = 1). These males were kept in the analyses reported in this study, as they were not visibly sick and had a developed brood pouch. Many females remained unmated in the female-biased treatments (F-L = 31, F-H = 21), but only two females were unmated under male bias (ML = 1, MH = 1).

Table 2. Summary of the number of individuals (N), density, operational sex ratio (OSR: start and end), fish length (mm), and density in the four treatments
   Density OSR endFish length
TreatmentSexN(fish/m2)OSR start(mean ± SE)(mean ± SE)
  1. *In 2008 one replicate had an extra male.

Low density      
Female-biased (F-L)Females100100.330.03±0.01191.75±2.43
 Males50 5  174.22±2.62
Male-biased (M-L)Females50 50.670.52±0.02192.78±3.55
 Males10010  173.42±1.81
High density      
Female-biased (F-H)Females100200.330.05±0.02190.79±2.31
 Males5010  173.16±2.59
Male-biased (M-H)Females50100.670.56±0.05189.40±3.42
 Males101*20 (21)*  173.24±1.90
Table 3. Mean ± SE of all measures (see Table 1) for each sex in each treatment. N = 5 for all measures. Mean and variance in reproductive (rs) and mating success (ms) are calculated across replicates
TreatmentSexmath formulamath formulamath formulamath formulaIIsmath formulamath formulasm
F-LFemales33.461095.421.551.891.10±0.250.85±0.160.99±0.050.90±0.100.42±0.060.37±0.06
 Males68.441181.193.232.350.26±0.040.24±0.060.57±0.150.30±0.090.22±0.060.15±0.05
M-LFemales82.521913.444.604.530.27±0.040.22±0.050.96±0.030.43±0.050.33±0.060.25±0.04
 Males40.611042.922.372.540.68±0.120.49±0.110.95±0.070.64±0.080.14±0.070.05±0.07
F-HFemales37.80124.001.701.470.83±0.150.53±0.040.86±0.080.63±0.090.35±0.030.18±0.06
 Males75.761168.813.431.380.21±0.010.12±0.020.45±0.250.15±0.080.21±0.030.13±0.05
M-HFemales85.122355.624.404.200.39±0.130.23±0.061.04±0.170.51±0.130.35±0.050.13±0.07
 Males42.47999.452.262.250.60±0.120.45±0.070.88±0.090.60±0.090.24±0.140.10±0.09

BATEMAN GRADIENTS (βss)

Female and male math formula were significantly positive in all treatments (Fig. 1, all P-values < 0.032). However, contrary to our predictions, the slope of the Bateman gradient did not differ between the two sex ratios or densities in females (Table 4 and Fig. 2). For males, the slopes of the math formula were steeper in the male-biased than female-biased treatments, but there was no effect of density (Table 4 and Fig. 2).

Table 4. The effect of sex ratio and density on the different measures of sexual selection sexes separately. All tests are two-way analyses of variance. Data on I and Is were square root transformed and math formula was squared before analysis
 OSRDensityOSR:density
 F-valueP-valueF-valueP-valueF-valueP-value
  1. Bold numbers indicate significant P values (P < 0.05).

math formula      
FemalesF1,17=0.980.348F1,17=0.0200.893  
MalesF1,17=7.570.014F1,17=0.0610.809  
I      
FemalesF1,17=23.31<0.001F1,17=0.080.79  
MalesF1,17=30.15<0.001F1,17=0.670.43  
Is      
FemalesF1,7=33.24<0.001F1,17=1.900.19  
MalesF1,17=19.94<0.001F1,17=1.470.24  
math formula      
FemalesF1,16=8.750.009F1,16=0.920.35F1,16=3.190.09
MalesF1,16=22.35<0.001F1,16=1.410.25  
s      
FemalesF1,17=0. 630.44F1,17=0.290.60  
MalesF1,17=0.100.76F1,17=0.310.59  
m      
FemalesF1,17=2.2890.15F1,17=7.620.01  
MalesF1,17=1.050.32F1,17=0.080.78  
Figure 1.

The Bateman gradient (math formula) for each sex in all four treatments (A) female-biased low density; (B) male-biased low density; (C) female-biased high density; and (D) male-biased high density. Females = open circles and solid lines, males = closed circles and dashed lines.

Figure 2.

Mean ± SE for the Bateman gradient (math formula) for each sex in all four treatments. Females = open circles, males = closed circles. N = 5 in all treatments.

In the female-bias low-density treatment, females had a steeper math formula than males but there was no difference in the math formula between the sexes in the female-biased high-density treatment (Table 5 and Figs. 1, 2). The male-biased treatments likewise exhibited no differences in the math formula between sexes (Table 5 and Figs. 1, 2).

Table 5. The effect of sex on the different measures of sexual selection in each treatment separately. All tests are one-way analyses of variance. Data on I and Is were square root transformed and math formula was squared before analysis
 F-LM-LF-HM-H
 F-valueP-valueF-valueP-valueF-valueP-valueF-valueP-value
  1. Bold numbers indicate significant P-values (P < 0.05).

math formulaF1,8=14.030.006F1,8=0.010.93F1,8=1.310.29F1,8=1.210.30
IF1,8=20.700.002F1,8=14.010.006F1,8=30.54<0.001F1,8=1.770.22
IsF1,8=16.920.003F1,8=4.990.06F1,8=87.62<0.001F1,8=7.350.03
math formulaF1,8=19.470.002F1,8=4.460.07F1,8=17.270.003F1,8=0.290.60
sF1,8=5.330.049F1,8=4.690.06F1,8=8.950.02F1,8=0.490.50
mF1,8=8.450.02F1,8=5.710.04F1,8=0.380.56F1,8=0.070.80

OPPORTUNITY FOR SELECTION (I) AND SEXUAL SELECTION (Is)

Both I and Is showed the predicted patterns in relation to sex ratio. For females, both measures were higher in the female-biased compared to male-biased sex ratio. Similarly, both measures were higher for males in the male-biased than female-biased sex ratio. Thus, the variance in both reproductive and mating success was higher for both sexes when they were the sex in excess. Density did not affect I or Is in females or males (Table 4 and Fig. 3A–D).

Figure 3.

Mean ± SE for (A) the opportunity for selection (I) for females; (B) I for males; (C) the opportunity for sexual selection (Is) for females; and (D) Is for males in all four treatments. Closed symbols represent estimates from the experiments whereas open circles represent estimates from the random mating simulations. Asterisks next to the observed values indicate significant differences between observed values and values obtained from random mating simulations. N = 5 in all treatments.

When investigating differences between sexes, females had significantly higher I and Is than males in both female-biased treatments (Table 5 and Fig. 3A, C). However, in the male-biased treatments I but not Is was significantly higher for males than females in the male-biased low-density treatment, and the opposite was the case in male-biased high-density treatment; Is but not I was significantly higher for males than females (Table 5 and Fig. 3B, D).

The observed values of I were only significantly higher than the simulated values under random mating for males in the male-biased low-density and female-biased high-density treatments (Table 6 and Fig. 3B). For females, the observed values of I were not higher than the simulated value of I in any treatment, but there was a trend toward a higher observed I in the female-biased low-density treatment (Table 6 and Fig. 3A). The observed values of Is, however, was significantly higher than simulated values for females under female bias but not under male bias (Table 6 and Fig. 3C). For males, the observed values of Is were not significantly higher than simulated values in any treatments except for a strong trend toward higher observed Is in the male-biased high-density treatment. In the female-biased high-density treatment the observed Is was actually significantly lower than Is under a random mating scenario (Table 6 and Fig. 3D).

Table 6. Results from one sample t-tests comparing observed values of I and Is to the mean values from the simulations for each sex in each treatment
 F-LM-LF-HM-H
 t-valueP-valuet-valueP-valuet-valueP-valuet-valueP-value
  1. Bold numbers indicate significant P-values (P < 0.05).

I
Females2.3950.075−1.0100.3691.6560.1731.4580.671
Males1.7590.1532.7870.0493.1380.0351.7560.154
Is        
Females3.3110.0291.2650.2753.1710.033−1.1330.321
Males0.3500.7441.7620.153−3.5500.0242.7210.053

THE MAXIMUM STANDARDIZED SEXUAL SELECTION DIFFERENTIAL (smax), SELECTION DIFFERENTIALS (s′), AND MATING DIFFERENTIALS (m′)

Both female and male math formula was affected by the sex ratio treatment in the predicted direction. When pooling the two densities math formula was 1.62 times higher for females in the female-biased than male-biased sex ratio, and 2.76 times higher in male-biased than female-biased sex ratios for males (Table 4 and Fig. 4A). This means that the maximum amount of precopulatory phenotypic evolution is higher for both sexes when they are the sex in excess. For females, there was also a trend toward an interaction between density and sex ratio and the difference between the sex ratios is slightly higher in the low-density treatments. However, there was no effect of density on male math formula (Table 4 and Fig. 4A).

Figure 4.

Mean ± SE for (A) the maximum standardized sexual selection differential (math formula); (B) the selection differential s′; and (C) the mating differential, m′ on body length for males and females in all four treatments. Dashed line indicates no selection on body length. N = 5 in all treatments.

Contrary to predictions, selection on body length as measured by s′ and m′ was not affected by sex ratio, either for females or for males. There was also no effect of the density treatment on male or female s′ or male m′ (Table 4 and Fig. 4B, C). However, in females we found a negative effect of the high-density treatment on m′. Thus, selection on body length with respect to mating was lower in high than in low density for females (Table 4 and Fig. 4C).

When investigating differences between the sexes in math formula, s′, and m′ in each treatment, we found that math formula and s′ both had higher values for females than males in the female-biased treatments. In fact, female precopulatory sexual selection (math formula) was 3.0 (low density) and 4.2 (high density) times higher than male math formula under female bias. However, sexes did not differ significantly in the male-biased treatments in math formula or s′, but there was a trend toward males having a higher math formula in the male-biased low-density treatment. Male math formula was 1.49 (low density) and 1.20 (high density) times higher than female math formula under male bias (Table 5 and Fig. 4A). Females tended to have a higher s′ than males in the male-biased low-density treatment (Table 5 and Fig. 4B). The mating differential (m′) was higher for females than males for both sex ratios in low density, but there were no differences between the sexes for m′ under high density (Table 5 and Fig. 4C).

Selection differentials, s′ and m′, calculated across replicates, were significant for females in all treatments except for m′ in the male-biased high-density treatment. For males, s′ was significant in both female-biased treatments, but s′ was not significantly different from 0 for males in the male-biased low-density treatment. Male m′ was only significantly different from 0 in the female-biased high-density treatment, but was also close to significant in the female-biased low-density treatment (Table 7 and Fig. 4B, C).

Table 7. Selection (s′) and mating differentials (m′) on body size across replicates with 95% confidence intervals (CIs)
TreatmentSexs95% CIm95% CI
  1. Bold numbers indicate values significantly different from zero (P < 0.05).

F-LFemales0.390.17, 0.650.350.16, 0.56
 Males0.200.07, 0.340.13−0.001, 0.28
M-LFemale0.300.17, 0.440.230.12, 0.33
 Males0.12−0.01, 0.260.05−0.08, 0.17
F-HFemales0.330.15, 0.510.170.05, 0.29
 Males0.180.04, 0.330.120.04, 0.22
M-HFemale0.320.13, 0.530.12−0.01, 0.27
 Males0.220.09, 0.380.09−0.02, 0.22

Discussion

The aim of this study was to investigate how sex ratio, density, and the interaction between them will affect the strength of sexual selection. In general, the measures of sexual selection used in this study behaved according to predictions in relation to sex ratio. Most measures of sexual selection were higher for females in the female-biased treatments and the opposite was true for males. For selection on body length, however, density but not sex ratio, influenced sexual selection (m′) on female body length. This finding demonstrates that increased population density can reduce selection on body length, presumably due to the breakdown of competitive mate-monopolization by females. In addition, we found that the opportunity for sexual selection, Is, was higher than expected under random mating for females under female bias. This indicates that sexual selection is operating on female S. typhle.

We expected the relationship between reproductive success and mating success to be steeper for females under female bias than male bias and vice versa for males. We found that our sex ratio treatment affected male but not female math formula. Thus, for the most competitive sex, the sex ratio manipulation did not change the potential for sexual selection in terms of fitness benefits from more matings. For males, the math formula were significantly steeper in the male-biased treatments, indicating a potential for sexual selection to be stronger on males under male bias than female bias. This is likely due to higher proportions of nonmated males under male bias than female bias. Male reproductive and mating success is limited by space in the brood pouch, whereas females are limited by their egg production (Berglund et al. 1989). Some females under male-bias benefited from unlimited access to males and mated with more males and had higher reproductive success than any male in any treatment, causing math formula to be as steep as under female bias. Nevertheless, math formula was positive in all treatments for both sexes demonstrating that there are significant fitness benefits of multiple mating for both sexes in this study.

We expected the sex ratio treatments to increase competition for mates among the sex in excess and therefore increase the variance in reproductive (I) and mating (Is) success. We found that both measures were higher for females in the female-biased treatments compared to male-biased treatments, and the opposite pattern in males. This result agrees with earlier studies that demonstrate that biased sex ratios have an effect on the genetic mating system in organisms such as mealworm beetles (Fairbairn and Wilby 2001), newts (Jones et al. 2004), broad-nosed pipefish (Jones et al. 2005), bank voles (Klemme et al. 2007), and marbled salamanders (Croshaw 2010). Within each treatment we also found significant differences in I between the sexes (higher for the sex in excess) for all treatments except for the male-biased high-density treatment. For Is, the sexes differed significantly in all treatments except for male-biased low-density treatment where there was only a tendency for males to have a higher Is. Overall, this suggests that the difference in variance between the sexes is generally stronger in the female-biased treatments. However, in this study we also simulated random mating to investigate how much of the observed variance in mating and reproductive success that could be attributed to random mating. Interestingly, the observed values of I were close to the random mating values in all cases for females, and only significantly higher for males in the male-biased low-density and the female-biased high-density treatments. However, observed values of Is were higher than the values obtained under random mating for females under female bias. This result confirms that sexual selection acts on females when the OSR is female biased in this species. A female-biased OSR is probably the most common state of this population of S. typhle in nature (Vincent et al. 1994). In the male-biased treatment, however, values of Is were not higher than expected from random mating for males nor females. Thus, the variance in the number of mates was not higher than expected by chance for females under male bias or males in any treatment. This corroborates our finding that the differences in I and Is between sexes is less pronounced in the male-biased treatments. Even if the male math formula, I, and Is responded to the sex ratio treatment, including I and Is under random mating suggests that males in this study seem to be under only weak sexual selection, also under male bias. Furthermore, the lack of difference between observed and random values implies that the variance in number of offspring is largely due to random processes and is therefore not expected to contribute to directional selection over time. Mating success is likely related to mating behavior in the experiment. However, the number of eggs received or transferred could be influenced by many factors such as interrupted matings and assessment of mate quality. Furthermore, the reproductive success of both sexes could be influenced by the proportion of eggs aborted or undeveloped. This could be affected by several factors such as female size, the number of eggs transferred, the potential for postcopulatory sexual selection, and male manipulation of offspring survival (Partridge et al. 2009; Paczolt and Jones 2010; Sagebakken et al. 2010, 2011; Mobley et al. 2011). All the mechanisms mentioned earlier could negate the effects of nonrandom mating success and lead to observed patterns of random variance in reproductive success.

The maximum potential for sexual selection on body length math formula (the upper limit to precopulatory sexual selection) was higher for both males and females when they were the sex in excess. Furthermore, we found that math formula differed between the sexes under female bias (higher for females) but not under male bias. Thus, our sex ratio treatment did not cause the maximum selection on body length to be any higher for males than females under male bias. However, we also measured direct selection (s′ and m′) on body length, a trait used in mate choice and in the formation of dominance hierarchies in this species, and found that it was not affected by our sex ratio treatment in the two sexes. As for math formula, our sex ratio treatment also did not cause selection on body length (s′) to be stronger for males than females under male bias. Nevertheless, positive selection on body length with respect to reproductive success (s′ significantly larger than 0) was found for both sexes in all treatments except for males in the male-biased low-density treatment. This result is expected, as both sexes exhibit a strong correlation between body length and fecundity in S. typhle (Berglund et al. 1986a; Ahnesjö 1992, 1995; Jones et al. 2005; Rispoli and Wilson 2008). Thus, selection differentials (s′) also measure fecundity selection whereas mating differentials (s′) are based on mating success as the measure of fitness and should therefore reflect precopulatory sexual selection (Jones 2009). In general, we found little evidence for sexual selection on male body length. If females actively choose to mate with larger partners or larger males are able to outcompete smaller males, we should have found significant male m′ in the male-biased treatments, where females should be choosy and males competitive due to an excess of males. However, larger males did not get more partners than smaller males in male-biased treatments. This is surprising, given that females generally prefer to mate with large males (Berglund et al. 1986a,1986b; Rosenqvist 1993), but see also also Sandvik et al. 2000). Our study suggests that females might not be choosy with respect to male length, or perhaps males do not form dominance hierarchies based on body size as females do (Berglund and Rosenqvist 2001, 2009). Because male-biased sex ratios are probably unusual in nature, except occasionally in the beginning of the breeding season (Vincent et al. 1994), female mate choice and male–male competition might not be common in the wild. The fact that we find weak selection on male body size with respect to mating success corroborates the results from the opportunity measures (I and Is), where in general the variance in reproductive and mating success where not higher than could be expected under random mating in both sex ratios for males. In contrast to the results for males, we did find significant sexual selection on females (nonzero m′) in all treatments except for the male-biased high-density treatment. However, large females still transferred more eggs than small females in this treatment (s′ was significantly different from 0), suggesting that larger females do not obtain more mates but transfer more eggs per mate. In all other treatments, larger females obtained more mates than smaller females, as expected.

Empirical evidence suggests that the effect of sex ratio on selection on body size is not uniform. Some studies have found stronger selection on length in the sex in excess (Höglund 1989; Souroukis and Cade 1993; Olsson et al. 2011). However, in common lizards high mating costs for females due to male harassment under male bias caused female choosiness to be stronger under female bias (Fitze and Le Galliard 2008). Thus, the relationship between sex ratio and choosiness is not as straightforward as predicted from classic sexual selection theory (Kokko and Monaghan 2001; Fitze and Le Galliard 2008).

The discrepancies between the opportunity measures (I and Is) and measures of selection on body length with regard to the effects of sex ratio could partly be explained by the fact that the variance in mating success and reproductive success could be generated solely by random mating, except for the variance in mating success for females under female bias. It is, however, surprising that selection on female length did not differ between the sex ratios for females. However, it is important to note that in this study we only investigated body length as the predictor of reproductive and mating success. It is possible that other morphological or behavioral traits are targeted by sexual selection.

We expected density to increase our measures of sexual selection especially under female bias due to high female–female competition and the potential for a few females to monopolize mating. However, most measures did not respond to our density treatment. Interestingly, we found that higher density had a negative effect on sexual selection on female body length (m′) regardless of the sex ratio. In the same experimental set up as this study, we also found that there were more intrasexual interactions in high-density than low-density populations (Aronsen et al. 2013). Female–female interactions increased as males became pregnant and the OSR became more female-biased in the female-biased low, but not high, density treatment. This suggests that female–female competition was high even when males were available for mating under high density when encounter rates with other females were high. In addition, the number of observed copulations was lower under high density suggesting more interrupted mating attempts (Aronsen et al. 2013). We suggest that decreased sexual selection on female body length is due to female–female competition being so intense under high density, that dominance hierarchies based on size suffered a breakdown, so the advantage of being a large female in terms of mating success is reduced.

Other studies on the effects of density on sexual selection on male characters have shown conflicting results from no change in large male advantage in the bluehead wrasse (Warner and Hoffman 1980) or increasing advantage with density for large males in pseudoscorpions (Zeh 1987). Furthermore, a decreasing advantage of large male traits at higher densities has been found in seed bugs (McLain 1992), fungus beetles (Conner 1989), and South African beetles (Knell and Pomfret 2008). These latter results are in line with our result on female m′. The similarities between our densities for most measures could be due to high encounter rates with potential mates in both density treatments, and in the future it would be interesting to investigate a broader range of densities, from very low to very high encounter rates.

Our study shows deviations from an earlier study on the effect of sex ratio on sexual selection in S. typhle conducted by Jones et al. (2005). They found that the male math formula was not significantly different from 0 in the female-biased treatment. In addition, they found no sexual selection on female body length (m′) in male- or female-biased sex ratios (Jones et al. 2005). The study by Jones et al. (2005) had fewer individuals per replicate, a more biased sex ratio (2:6 in the male-biased and female-biased treatments) and individuals were only allowed to breed for 72 h. All these factors could perhaps have caused different results from this study. Alternatively, this study could simply have more statistical power to find a significant relationship. This is most likely the explanation for the significant female m′ in both sex ratios in our study in contrast to the results of Jones et al. (2005).

In this study, we chose to end the trials based on mating activity instead of standardizing the length of each trial. This could have weakened the strength of sexual selection on males in female-biased treatments as nearly all males mated in these treatments. There would have been more unmated individuals in some replicates if we had ended the trial after a fixed number of days or hours. However, we believe that standardizing by time could have caused a large amount of variance between replicates due to differences in the speed of mating which is likely influenced by variation in temperature and eagerness to mate.

In summary, accumulating evidence suggests that demographic factors such as sex ratios and population density have a strong influence on local mating dynamics and may shape the mating system of different populations and species. Our results indicate that both sex ratio and density differences are likely candidates to explain part of the variation in sexual selection and mating systems between and within populations that can be observed in nature. We believe our results are relevant also to species with conventional sex roles, where males are the sex under stronger sexual selection. Unique to this study, we applied several quantitative measures of sexual selection derived from the genetic mating system, including selection on a morphological trait important to mating decisions and reproductive success in this species. We encourage more empirical studies that adopt a similarly wide-scale experimental approach to disentangle the effects of specific ecological and demographic factors that affect mating systems and sexual selection. The ultimate goal is to achieve a more complete understanding of how sexual selection operates in nature.

Associate Editor: R. Bonduriansky

ACKNOWLEDGMENTS

The authors thank J. Sundin, A. M. Billing, L. Cats Myhre, G. Sagbakken, E. Berglund, H. Höglund, D. Höglund, R. Höglund, and R. A. Aronsen for field assistance. We also thank A.-L. Olsen and R. Höglund for assistance in the lab, F. Fossøy for Genemapper advice and A. Jones, S. Wacker, and C. Pelabon for valuable discussions. Funding was provided by the Norwegian Research Council 186163/V40 (GR), the Swedish Research Council (AB), the Association of European Marine Biological Laboratories (GR), Nordic Marine Academy (TA), and the National Science Foundation (KBM). All handling of fish was done under license Dnr 118-2008 from the Swedish Board of Agriculture.

Ancillary