MALE–FEMALE COEVOLUTION IN THE WILD: EVIDENCE FROM A TIME SERIES IN ARTEMIA FRANCISCANA

Authors


Abstract

Sexual conflicts are ubiquitous in nature and are expected to lead to an antagonistic coevolution between the sexes. This coevolutionary process is driven by selection on sexually antagonistic traits that can either be directional or fluctuating. In this study, we used dormant cysts of Artemia franciscana, collected in the same population in three different years over a 23-year period (corresponding to ∼160 generations in this system), to investigate male–female coevolution in natural conditions over time. We performed a cross experiment study where reproduction of females mated to males from the past, present, or future was monitored until death. In agreement with a model of “fluctuating selection,” we found that females survived better and had longer interbrood intervals when mated with their contemporary males compared to when mated with males from the future or the past. However, female weekly and lifetime reproductive successes displayed no differences between contemporary and noncontemporary matings. Finally, the coevolutionary patterns (“arms race dynamics” or “fluctuating selection dynamics”) possibly acting on female relative fitness could not be discriminated. This study is the first direct demonstration that the process of male–female coevolution, previously revealed by experimental evolution in laboratory artificial conditions, can occur in nature on a short evolutionary time scale.

Sexual conflict occurs when the fitness optimum of a trait differs between males and females (e.g., Parker 1979). Traits selected in opposite directions in both sexes can result in an “evolutionary chase” (Parker 1979). A large number of species has actually been shown to exhibit antagonistic coevolution between males and females in reproductive behaviour, morphology, or physiology (Arnqvist and Rowe 2005a; Holland and Rice 1998). For instance, in polyandrous species each male is selected to maximize his mate's resource allocation toward its own offspring (because offspring produced later in life are likely to be fathered by other males, Rice 2000). Hence, males may “manipulate” females to increase their short-term fecundity. In contrast, females are selected to maximize their lifetime reproductive success and to resist male mating persistence (Rice 2000). Female “manipulation” by males has been exemplified in Drosophila, where males increase their own fitness by manipulating the egg-laying rate of their mate at the expense of female survival (Chapman et al. 1995; Chen et al. 1988; Fowler and Partridge 1989). Differences over mating rate optima between sexes have also been found in other insect species (Arnqvist and Nilsson 2000) and similar conflicts of interests between males and females are in fact ubiquitous in nature (Holland and Rice 1998).

Antagonistic coevolutions are driven by a variety of processes that involve reciprocal adaptive genetic changes between species—for example in host–parasite systems (Woolhouse et al. 2002)—or between sexes—in species with sexual conflicts (Arnqvist and Rowe 2005a). Among these processes, arms race dynamics (ARD) and fluctuating selection dynamics (FSD) are usually opposed (Arnqvist and Rowe 2005a; Gandon et al. 2008; Woolhouse et al. 2002). Under arms race selection, the phenotypic traits under conflicts are altered unidirectionally over time. This leads to constant “phenotypic innovations,” which occur through the selection of new combinations of alleles or genes. For instance, grasping and anti-grasping structures in male and female water striders are supposed to have evolved under ARD (Arnqvist and Rowe 2002a; Arnqvist and Rowe 2005a). Under fluctuating selection the phenotypic traits under conflicts are oscillating periodically over time. This “phenotypic repetition” occurs either through the recruitment of new alleles or genes or through frequency changes of pre-existing alleles or genes over time (e.g., through negative frequency-dependent selection). This latter case can lead to a stable polymorphism, for example it is thought to drive the maintenance of female color polymorphism in damselflies (Andres et al. 2002; Arnqvist and Rowe 2005a).

Sexual conflicts have been intensively investigated over the last decade; yet, studies have mainly focused on insect behavior and unambiguous experimental data on coevolutionary processes in other taxa remain scarce (Tregenza et al. 2006). Because sexual coevolution is a dynamic process, several authors (e.g., Chapman and Partridge 1996; Rice 1996, 2000) have hypothesized that contemporary conflicts may be hidden by a long history of adaptations and counteradaptations. Several studies have tried to circumvent this problem using two main approaches. First, experimental evolution and genetic manipulation have been used to impede the adaptation of one sex to the other (Holland and Rice 1999; Rice 1996). In a study on Drosophila, Rice (1996) showed that males could rapidly adapt to a static female phenotype, when females were prevented from coevolving with males. However, even if such studies show that there is scope for sexual conflicts to occur, they most often involve complex experimental setups that may involve artificial selective conditions. Overall, these experiments cannot inform us on the direction and strength of similar coevolution in natural populations. Second, experiments involving crosses between individuals from different populations have been used to reveal coevolution between sexes. Indeed, males and females living in different populations can express coevolving characters, which follow divergent evolutionary trajectories (Parker and Partridge 1998). In these studies, spatial variation between populations is thought to mirror variation through evolutionary time within a population and, when comparing crosses of females mated to males from different populations (including from their own), females should be more resistant to manipulative traits of males with which they coevolved (Parker and Partridge 1998). Many experiments have been conducted to test this prediction (Andres and Arnqvist 2001; Brown and Eady 2001; Nilsson et al. 2002). However, up to now, the results from these studies are partly inconsistent (Chapman et al. 2003). One inherent problem to this approach is that only pairwise population comparisons can be performed, which in a factorial experiment prevents the combination of the information from all the crosses to infer the coevolutionary process.

Ideally, sexual conflicts should be investigated with a historical perspective (Holland and Rice 1998; Pizzari and Snook 2003), for example studying a population over a large temporal scale (comparing individuals at time t with their ancestors over many generations). However, the model organisms currently used to investigate such conflicts (Drosophila and more generally insects) do not allow the comparison of live individuals several generations apart, preventing the direct study of coevolution between sexes. In the last decade, some attempts have been made to study such long-term (co-)evolution in host–parasite systems through time-shift experiments (Gaba and Ebert 2009). These experiments compared the proportion of infected hosts when confronted with “past,”“contemporary,” and “future” parasites (Buckling and Rainey 2002; Decaestecker et al. 2007). However, the organisms used in this host–parasite literature reproduce mostly (or totally) asexually, preventing the investigation of male–female coevolution in these systems.

In this article, we used the brine shrimp (Artemia franciscana) to study long-term coevolution between sexes in a natural population. The biology of Artemia, makes it an ideal model to directly reveal sexual coevolution in natura. Artemia species reproduce either ovoviviparously or oviparously. The latter mode is favored by either high salinity, high temperature, low oxygen concentration, food shortage, or short photoperiods (Clegg and Trotman 2002; Nambu et al. 2004). In this case, they produce encysted embryos called cysts (Clegg and Trotman 2002), that can be stored at low temperature for long periods of time before hatching and initiating the next generation (Lavens and Sorgeloos 1987). Thus, multigenerational experiments with controlled and synchronized hatching are possible in this species (Reznick 1993).

In this study, we tracked male–female coevolution over a period of 23 years (ca 160 generations, see below), using dormant cysts from a wild population of A. franciscana from the Great Salt Lake (Utah). More specifically, we used newly produced cysts, collected in three different years (1985, 1996, and 2007) in a cross experiment where females were mated either with males from their own year or from other years. We measured different life-history traits such as female lifetime survival and reproductive output. We investigated male–female coevolution patterns by testing specific statistical interactions between male and female year effects. Ultimately, we examined whether fluctuating or directional selection is driving male–female coevolution in this species.

Materials and Methods

MODEL SPECIES A. FRANCISCANA

Artemia species, also known as brine shrimps (Pancrustacea, Banchiopoda, Anostraca), have a worldwide distribution and are found in inland salt lakes, coastal lagoons, and solar saltworks (Lenz and Browne 1991). The environment of this hypersaline species can easily be recreated in the laboratory (Lenz and Browne 1991). In particular, the most widespread member of the group, the bisexual A. franciscana, has been the focus of a number of life-history studies (Browne 1980, 1982; Browne et al. 1988, 2002; Browne and Wanigasekera 2000; Shirdhankar et al. 2004). In this polyandrous species, mate guarding occurs and male antennae develop into claspers that males use to clasp females during mating (Lochhead 1950). This clasping usually lasts longer than just the time needed for copulation (Wolfe 1973); thus, clasping is probably costly for the female (e.g., in term of foraging activity).

This study was focused on one population of A. franciscana from the Great Salt Lake (Utah). Dormant cysts sampled in three different years (June 1985, February 1996, and September 2007), spanning a 23-year period, were kept refrigerated at the Artemia Reference Center (Ghent, Belgium). Given the high population size (Ne > 109, Wurtsbaugh and Maciej Gliwicz 2001), the homogeneity of environmental conditions in the South part of the lake (where the cysts were collected) and the mixing effect of winds and currents, we expect negligible neutral differentiation between our samples. First reproduction appears in late April in this population and mature females’ density increases throughout spring and summer to peak in early October (with a shift from 0% to 96% of oviparously reproducing females over this period, Wurtsbaugh and Maciej Gliwicz 2001). Considering an average generation time of 23 days at 15°C (von Hentig 1971), we estimated an average of seven generations per year over this 6-month period (±4 generations, based on the data at 10°C and 20°C). Hence, we estimated that our experiment spans approximately 160 generations (±90 generations).

CULTURE CONDITIONS

Nauplii and adults were reared in concentrated brine water (Thalasea ©, Salins du midi, Aigues-Mortes, France) diluted to a 40 g/L salinity with dechlorinated tap water. Individuals were kept at 25 ± 1°C under constant fluorescent lighting and fed with Dunaliella salina and D. tertiolecta algae. These algae were cultured and frozen in the laboratory, and subsequently distributed four times a week after defrosting, providing a daily individual amount of around 1.6 million cells (about half of the adult maximum ingestion rate, Reeve 1963). Culture medium was replaced weekly. Cyst decapsulation and hatching protocols were modified from Bengtson et al (1991). Cysts were sieved on a 120-μm mesh, decapsulated with a brief exposure (< 10 min) to a sodium hypochlorite solution (2.6%), and rinsed with fresh water. Decapsulated cysts were incubated in a 5 g/L salinity medium buffered with 2 g/L of NaHCO3 (pH = 8.3). We used cylindroconical hatching containers (3 L) where aeration was applied from the bottom with an air-water-lift system. After emergence, first-instar nauplii were placed in large containers (8 L) and fed daily. This procedure was performed independently for the three different years of cyst origin.

Before individuals reached sexual maturity, their sex was assigned on the basis of sexual dimorphism (Lochhead 1950). After maturity, single pairs of males and females from the same year were placed in 0.5-L pints to produce an F1 generation. Each brood of nauplii was isolated from their parents 24 h after the first nauplii were observed. Within two weeks after brood isolation, individuals were isolated in test tubes (2.5 × 7 cm) in 30 trays containing up to 70 tubes. Tubes had a filter bottom to allow the water to flow (mesh size: 120 μm). Each brood was identified and each tray contained from one to six families (randomized across years), with no more than nine siblings from a given family. Tray position on the laboratory shelf was weekly randomized to avoid local shelf effects.

EXPERIMENTAL SETTING

The number of families used was balanced across the three years (13 in 1985, 14 in 1996, and 14 in 2007). Female survival and reproduction were monitored individually, whereas males were randomized within each given year. Hence, when the F1 generation was old enough to be sexed, males from the same year were pooled together in large containers whereas females were kept individually isolated in test tubes. A total of 763 females were followed (281 in 1985, 290 in 1996, and 192 in 2007). Sibling females were assigned to one of three time-shift treatments (i.e., male from the past, present, future, see below). Male years were represented in each family except for two families of low sample size, and females were paired before their sexual maturity, either with a male from 1985, 1996, or 2007. The three male treatments were equally represented among sibling females on the same tray.

Because all the males used in the experiment did not reach sexual maturity at the same time and to average male effect among years, males were replaced once a week during all the term of the experiment, females were then randomly allocated a new male from the same treatment year. Males clasping females could not be removed because of possible harm to them or their mate. Hence some females remained with the same male for consecutive weeks. On average females remained ∼8 days with the same male. Dead males were immediately replaced.

Survival (longevity) and reproductive parameters (laying date of consecutive clutches, type of offspring (cysts/nauplii), clutch or brood size) were recorded three times a week for each female until its death (last female died at 203 days). Clutch or brood size was estimated based on photographs (repeatability 89% for nauplii and 98% for cysts, repeatability estimations based on Lessells and Boag 1987). Pictures were taken using a Pixelink (PL-A662) high-resolution color camera.

STATISTICAL ANALYSES

We analyzed a suite of female life-history traits: survival (time to death), lifetime reproductive success (LRS, lifetime number of offspring), weekly reproductive success (WRS, number of offspring in consecutive weeks), reproductive rate (interbrood interval), relative fitness (a combination of those traits that we define below). Male–female coevolution was investigated on the suite of traits illustrated in Fig. 1. A significant interaction term (FemYr × MaYr) after accounting for female and male years of origin (FemYr + MaYr) is an indication of such coevolution. However, simpler models can reach a better compromise between number of parameters and deviance (i.e., explanatory power). They are also more insightful, because they allow the inference (and prediction) of specific coevolutionary patterns. Hence, we used different functions of time-shift (ΔT= male year – female year) to test specific male–female coevolution patterns (Fig. 2). We directly fitted ΔT for directional effects (e.g., arm race dynamics, hereafter ARD, Fig. 2A) and a combination of ΔT and ΔT2, for more complex shapes (e.g., FSD, Fig. 2B,C). Finally we also used δ(ΔT), where δ(·) stands for the Dirac delta function, to model contemporary versus all noncontemporary matings (e.g., rapid fluctuating coevolution, Fig. 2D). It is important to note that the model including ΔT along with male and female year effects is undefined (unlike ΔT2, or δ(ΔT)). Thus, it could only be fitted with either of these variables, which may be problematic if both year effects are large. In addition, a set of potential confounding effects were also included in all the analyses: number of siblings of the focal female (SibSize), female birth date (FBD), age of female when the first male was introduced (AFM), the duration the female spent alone between sibling isolation and male first introduction (Iso, see Fig. 1), tray effect (Tray), and family effect (Sib, included in the analyses as a random effect, unless otherwise stated). All analyses were performed using R (version 2.10.0). We detail below specific models used for each trait. Model selection was based on corrected Akaike's information criterion (AICc; Hurvich and Tsai 1989). Whenever a model with a function of time-shift provided a better fit than the additive model (ΔAICc < 2), the percentage of the residual deviance (R2_Dev) from the additive model accounted for by this model was calculated (Grosbois et al. 2008; Skalski 1996).

Figure 1.

Schematic illustration of the survival and reproductive traits analyzed along with some confounding covariates. Female lifetime reproductive success (LRS) of females B, C, and D differs from the reference female A (with constant brood size and interbrood interval) in one trait: lifespan (B), weekly reproductive success (WRS, C) and interbrood interval (reproductive rate, D). Tick marks on the horizontal axis represent the weeks after first reproduction. Vertical arrows are observations. Light gray rectangles represent female life span. Dark gray circles represent reproductive events and circle size represents brood/clutch size. WRS is the number of offspring produced over a 7-day period. LRS is the sum of WRSs. Female relative fitness could not be represented. FBD = time between the start of the experiment and female birth. AFM = female age when the first male was introduced. Iso = isolation time between brood individualization and first male introduction.

Figure 2.

Expected time-shift effect on female fitness under different coevolutionary processes. Time-shift (ΔT= male year – female year). Male–female coevolution was tested with different interaction shapes. Female fitness under “arms race” selection (ARD model) is altered unidirectionally with time-shift. (A) Linear time-shift effect (ΔT). Female fitness under “fluctuating selection” (FSD models) is altered with absolute time-shift. (B and C). Quadratic effects (ΔT2T and ΔT2), D. different coefficients for contemporary and noncontemporary matings (δ(ΔT)). The experiment comprises different male–female combinations for the different time-shifts (3 for ΔT= 0 year, 2 for |ΔT|= 11 years and 1 for |ΔT|= 22 years).

FEMALE SURVIVAL

Female survival was analyzed using the survival package in R. This analysis was aimed at testing the effect of male time-shift on female survivorship and focused on the number of days between male introduction and female death. An interval censored model with the age of last observation alive and the age at first observation dead was used (Gómez et al. 2009). A small number of females had unknown date of death or could not be assigned a mate because of male shortage during the experiment. These females (24 and 18 respectively) were right-censored to the last valid observation (RC females hereafter). First, the full model was fitted using different survival curves available in the survival package (Weibull, Extreme value, Exponential, Gaussian, Logistic, Lognormal, Loglogistic). Because the Weibull was, by far, the best model (ΔAICc >> 2 for other models), it was used in all subsequent model selection. The Weibull model allows age-specific mortality rate (α) to increase or decrease with age (Crawley 1993), in addition to the fit of the survival function scale (λ). The package allows fitting only one factor to α. We tested either female year or discrete time-shift effects. Family was specified as a random effect (frailty option, Therneau et al. 2003).

FEMALE LIFETIME REPRODUCTIVE SUCCESS

Because female lifetime reproductive success (LRS) was clearly bimodal with many nonreproducing females (30%), we analyzed separately the probability to reproduce (logistic regression, lme4 package) and LRS of reproducing females (negative binomial count model with the glmmADMB package). In the former, tray and family were specified as random effects. In the latter, the proportion of nauplii (Prop) (as opposed to cysts) was added to the set of variables previously mentioned to control for different costs of cysts versus nauplii production. RC females were excluded from these analyses.

FEMALE WEEKLY REPRODUCTIVE SUCCESS

Female Weekly Reproductive Success (WRS) was computed as the number of offspring (cysts and nauplii) produced per week after the first reproduction. To analyze the variation of reproductive effort with age, we regressed WRS on Time (number of weeks of reproduction) and Time2 to allow for a possible nonlinear effect (along with all other covariables mentioned previously). WRS was analyzed with a negative binomial error in the glmmADMB package. To account for nonindependence of consecutive WRS values for a given female, female identity was included as a random effect. Nonreproducing females were excluded from this analysis. Only the data to the last valid observation were used for RC females.

FEMALE REPRODUCTIVE RATE

We analyzed female interbrood interval following an approach similar to a survival analysis where each reproductive event is analogous to a death event. This method is equivalent to a study of mortality rate in different populations. Each female here is comparable to a population and the number of clutch/brood corresponds to the number of individuals in each population, hence the different reproduction events are equivalent to the death of an individual. For this analysis, the occurrence of reproduction between Time minus one week (Time – 1) and Time was analyzed with an interval censored survival model (Gómez et al. 2009). We used one-week intervals to circumvent problems that would occur for smaller or larger time intervals. A much larger interval would frequently merge two consecutive reproductive events (they occur on average every 5–10 days in our experiment). On the other hand, a much smaller interval would create a negative correlation between the probabilities of observing a clutch/brood in consecutive intervals, because minimal interbrood interval is ∼4 days in Artemia. Similarly to our survival analysis, we tried here different “survival” curves in the survival package and the Weibull was the best to fit the data (ΔAICc > 2 for other curves). Only the data to the last valid observation were used for RC females.

FEMALE RELATIVE FITNESS

To combine survival and the rate of offspring production in a synthetic measure of fitness, we used the intrinsic rate of increase r, which represents the growth rate of a population initiated from a single individual, once the stable age structure has been reached (Charlesworth 1994; Peters et al. 2003; Vassilieva and Lynch 1999). It was computed only for reproducing females, using the WRS dataset, as the largest root r of

image(1)

where l(x)m(x) is the product of survivorship to and productivity (WRS) at week x. Relative fitness was measured by dividing each female r by the mean r. This variable was analyzed with a gamma error using the stats package. Family (Sib) was included as a fixed effect in the model (this package does not allow random effect specification).

Results

We investigated whether the life-history traits of sibling females differed in the presence of a male from their past, present, or future. A vast majority of females (95%) were followed until death and the experiment lasted 203 days. We detail below, for each female trait separately, the results from the best models. In Tables S1–S6, all models with AICc differences lower than 10 compared to the best model are presented (Anderson et al. 2001).

FEMALE SURVIVAL

Average female lifetime after male introduction was 29.6 days (SD = 27.2 days, median = 21 days, min = 1 day, max = 158 days, excluding RC females). A strong effect of female year of origin was observed during the experiment with increased survival for 2007 females compared to 1985 and 1996 females (Fig. S1). Models including ΔT2 and δ(ΔT) effects (i.e., corresponding to gradual or rapid fluctuating selection, FSD) ranked best with ΔAICc lower than two (Table S1). Fitted survival curves were qualitatively equivalent in these models. Estimates from the ΔT2 model are presented in Table 1. This model accounted for 94% of the residual deviance between the additive and saturated models (Table S1). The fitted ΔT2 effect was negative, indicating that overall females survived best when mated with contemporary males (Fig. 3 and S2). According to this model, females with a time-shift of ±11 and ±22 years survived on average 3% and 12% less, when compared to those mated with contemporary males. The Weibull model also indicated that, for the three years, female age-specific mortality rate is increasing with age in this species (Log(1/α) < 0, Table 1).

Table 1.  Parameter estimates from the best survival model. This model included the male and female year effects (FemYr and MaYr) and the ΔT2 interaction, indicating fluctuating selection. Other variables accounted for potential confounding effects in the model. Estimates are given on a log-scale relative to the intercept (i.e., average Weibull scale of 85 male–female combinations raised in Tray 1). The negative ΔT2 estimate indicates that female survived best when mated with contemporary males (see Fig. 3). Negative 1/α estimates fitted for the different female years revealed an increase of age-specific mortality over time.
Variable EstimateSE
  1. 1A total of 29 estimates for specific Tray effects not shown.

  2. 2Different senescence parameters were fitted for each female year in the model (see Methods).

  3. 3The standard error associated with the random effect variance was not provided in the package.

Intercept 3.260.62
FemYr 96 0.310.51
FemYr 07 1.550.53
MaYr96−0.110.05
MaYr07−0.030.05
ΔT2−0.030.02
FBD 0.120.14
AFM 0.230.15
Iso−0.280.13
SibSize 0.010.01
SibSize:FemYr96−0.010.01
SibSize:FemYr07−0.010.01
Tray1 −
Log(1/αFemYr85)−0.730.05
Log(1/αFemYr96)−0.500.05
Log(1/αFemYr07)2−0.690.07
Random(Sib) 0.183
Figure 3.

Fitted effect of time-shift on female survival. The best model indicated a decrease of female survival when mated to noncontemporary males. Errors bars denote 95% confidence intervals based on bootstrap values calculated with the standard error of the ΔT2 coefficient (Table 1). The other best models (with ΔAICc < 2) gave qualitatively similar results.

FEMALE LRS

The proportion of nonreproducing females was variable across years (0.3, 0.4, and 0.06 in females from 1985, 1996, and 2007, respectively). The ΔT, ΔT2, and additive (i.e., male and female effects but no male–female interaction) models fitted best the data on the probability to reproduce (ΔAICc < 2, Table S2). Thus, we did not detect a clear effect of time-shift on the probability to reproduce. Reproducing females produced 135.3 offspring on average (SD = 138.2, median = 86, min = 6, max = 860, excluding RC females) over their lifetime. The additive (with female effect only), δ(ΔT), ΔT2, and ΔT models fitted the LRS data best (Table S3). Thus, again, we did not find any simple coevolutionnary pattern. Interestingly, female LRS increased with the proportion of nauplii (Table 2 and Table S3), which may suggest either that nauplii are cheaper to produce than cysts or that females in good condition tend to produce more nauplii.

Table 2.  Parameter estimates from the best WRS model. Estimates are given on a log-scale relative to the intercept. Time and Time2 estimates indicate an increase of female early Weekly Reproductive Success. The positive interaction estimate Time: ΔT2 indicates that female tended to produce less offspring per week when mated with contemporary males.
Variable EstimateSE
  1. 1A total of 29 estimates for specific Tray effects not shown.

Intercept 3.5340.106
Time 0.0290.013
Time: ΔT2 0.0030.002
Time2−0.0020.001
SibSize−0.0020.001
Iso 0.0760.017
Tray1
Random(Sib) 0.0070.006
Dispersion parameter 1.1550.039

FEMALE WRS

Reproducing females produced on average 31.6 offspring per week (SD = 26.6, median = 28, min = 0, max = 169) during each week of their lifetime. All best models included a quadratic effect of the number of weeks after first reproduction (Time+Time2,ΔAICc < 2, Table S4). However, complementary analyses showed that this pattern was due to an increase of WRS early in life, followed by a plateau, rather than to true senescence. In addition, some of these models included an effect of time-shift (ΔT2 or ΔT) on the intercept and/or an interaction between Time or Time2 and a function of time-shift (i.e., either ΔT2, ΔT or δ(ΔT), Table S4). The fitted values of the intercept and the interaction with Time+Time2 of ΔT were negative, whereas those fitted for ΔT2 and δ(ΔT) were positive (data not shown). Moreover, the additive model (without any time-shift term) was among the best models (Table S4). Thus, we did not detect any consistent effect of time-shift on the age-specific WRS.

FEMALE REPRODUCTIVE RATE

The interaction, FSD (ΔT2 and δ(ΔT)) and additive models fitted Weibull scale parameter to the reproductive rate data similarly (ΔAICc <2, Table S5). As for female WRS, we did not detect a consistent effect of time-shift on this parameter. However, all these models fitted different Weibull shapes for contemporary and noncontemporary matings, demonstrating a difference in age-specific reproductive rate between both treatments (Table S5). The two first best models accounted for only 14% and 10% of the residual deviance between the additive and saturated models (Table S5). Fitted Weibull shape was lower when females were mated to contemporary males, indicating that reproduction rate was lower in this treatment (Fig. 4). In addition, the fitted Weibull shapes were higher than one in both treatments suggesting that, on average, reproduction rate increases with age in this species (both curves from females with contemporary and non contemporary males lie above the curve of a reference female with constant interbrood interval in Fig. 4). Consequently, the decrease of interbrood interval was sharper when females were mated to noncontemporary males (Fig. 4).

Figure 4.

Comparison of fitted reproductive rate between contemporary and noncontemporary matings. The best models suggest that females had a lower egg-laying rate when mated with contemporary males, compared to the females mated with noncontemporary males. Interbrood interval decreases with age for females from both treatments. Λ= 0.79 (Weibull Scale), α(ΔT=0))= 1.34, α(ΔT≠0))= 1.42, and αref= 1 (Weibull Shapes).

RELATIVE FITNESS

Mean female relative fitness was 1 (SD = 26.6) and ranged from 0.48 to 2.78. The ΔT2+ΔT and ΔT models fitted the data best (ΔAICc < 2, Table S6). Both these models indicated that female fitness decreased with time-shift (Fig. 5 and S3). The ΔT2+ΔT model accounted only for 22% of the residual deviance between the additive and saturated models (Table S6). As a consequence, unambiguous discrimination between ARD and FSD patterns on this trait was not possible.

Figure 5.

Fitted fitness relative change compared to contemporary pairs. The ΔT2+ΔT and ΔT models indicated a decrease of female relative fitness with time-shift. Errors bars denote 95% confidence intervals based on bootstrap values calculated with the standard error of the time-shift coefficients of the models.

CONSISTENCY ACROSS TRAITS

Fig. 6 pictures the main results of the study. Female survival and interbrood interval were affected by ΔT2 and δ(ΔT), respectively, which suggests that these traits are under fluctuating selection (Fig. 6). On the contrary, we did not detect any clear trend on female lifetime and WRS (Fig. 6). Unfortunately, the selection pattern on female relative fitness could not be clearly identified.

Figure 6.

Schematic illustration of the results from the study. Females survived best when mated to contemporary males (rectangle length in A, B and C). Female lifetime reproductive success did not differ between time-shifts (same number of circles in A, B and C). Female interbrood interval decrease was shallower when females were mated to contemporary males (longer distances between the circles in A, compared to B and C). Note that even if female weekly reproductive successes seem to differ between A and B and C, no difference was found for this trait between time-shifts. Symbols are described in Figure 1.

Discussion

In this study, we investigated male–female coevolution through time using dormant cysts of A. franciscana collected in the same population in different years (1985, 1996, and 2007). This is the first time sexual conflicts in natura are investigated over this time scale (spanning about 160 generations). Interestingly, our results suggest that male–female coevolution occurs in this population. The analyses of survival and reproductive rate patterns are consistent with fluctuating selection, whereas analyses of female relative fitness failed to distinguish arms race from fluctuating selection.

FEMALE YEAR EFFECT ON SURVIVAL AND REPRODUCTION

Our experiment revealed that all life-history traits investigated, showed significant differences across females from different years of origin (except for WRS). Analyses of survival and LRS showed that females from 2007 lived longer and produced more offspring in total than females from 1985 and 1996. Interestingly this effect was not found in the WRS analysis. This discrepancy indicates that the number of offspring per clutch/brood was equivalent across years, although female survival and interbrood interval were different.

These strong year effects could be either due to a genetic change of the population or to different environmental conditions (e.g., time of conservation), which would have been passed on to the focal F1 offspring in our experiment through maternal effects. To our knowledge the effect of conservation time on life-history traits has never been investigated in Artemia. Further investigations rearing several generations in laboratory conditions would help disentangle genetic from environmental factors acting on this strong year-of-origin effect.

SHAPE OF MALE–FEMALE COEVOLUTION IN ARTEMIA

In this study, we found that female survival and reproductive traits followed patterns consistent with male–female coevolution in a wild population of A. franciscana. Earlier studies have investigated sexual conflicts between species (e.g., Arnqvist and Rowe 2002b) or between populations within species (e.g., Andres and Arnqvist 2001; Brown and Eady 2001; Nilsson et al. 2002), but to our knowledge, this is the first study suggesting such a coevolutionary pattern in a natural population over such a time scale.

Different variables were used to test alternative coevolutionary scenarios (see Methods). Our results indicated a time-shift effect on some of the traits. Analyses of survival and reproduction patterns indicated FSD scenarios as very likely. Disentangling the FSD from the ARD scenario was however statistically impossible for female relative fitness.

The clearer statistical effects of time-shift concerned female survival and reproductive rate (Fig. 3 and 4). Survival was the highest when females were mated to contemporary males and decreased by 3% and 12% when mated to males with 11-year and 22-year time-shifts (Fig. 3). In addition, the decrease of interbrood interval with age was the highest, when females were mated with noncontemporary males (Fig. 4). In theory, this time-shift effect on female interbrood interval should translate into an interaction between time-shift and Time in the WRS analysis (Fig. 6). However, evidence for such an interaction was mitigated, because the additive model (without any time-shift effect) was among the best WRS models (Table S4). We suppose that the effect of time-shift on interbrood interval was too small to be detected in the WRS analysis. In addition, we observed an increase in female WRS over the first seven weeks following first reproduction, after which WRS plateaus. Finally, we did not find any clear effect of time-shift on female LRS. This result may not be surprising as the increased survival of females mated with contemporary males tended to be counterbalanced by the shallower decrease in interbrood interval. It simply reflects trade-offs between survival and reproduction. Such a trade-off has already been shown in A. franciscana (Browne 1982). Its presence makes it difficult to determine the primary fitness trait influenced by male behavior across time-shifts.

Interestingly, we found a decrease in female fitness with increasing time-shift. However ARD and FSD selection patterns on female relative fitness could not be statistically disentangled (Fig. 5). This result contrasts sharply with those obtained for the other traits (survival, LRS, WRS and reproductive rate), where females with noncontemporary males (i.e., with positive or negative time-shifts) behaved similarly. Because female relative fitness was computed as a combination of survival and reproduction, we also expected that trait to follow a similar pattern. These differences occur because early reproduction weighs more in the fitness measure than late reproduction (an effect not taken into account in the LRS analysis). Consistent with this explanation, we found an effect of ΔT on WRS’ intercept in some of the best models in Table S4. In all cases, the fitted ΔT effects were negative, which indicates that WRS early in life was higher for females with negative time-shifts. Discriminating among directional and fluctuating selection patterns acting on female fitness would require sampling over a wider time scale. This would also allow to ascertain the occurrence of the coevolutionary cycles observed on female survival and interbrood interval and would help to better characterize the dynamics (e.g., period and phase) of these cycles. Nevertheless, although ancient cysts are readily available for such experiments, their hatching remains a challenge.

Furthermore, although our experiments support the hypothesis that female survival and reproduction are affected by males, the underlying mechanisms mediating the male–female interaction remain unknown. Indeed, the persistence of a sexual conflict requires divergent selection on one or several traits involved in the male–female interaction (Rowe and Day 2006). However, finding out which traits generate the conflict is not straightforward. Some hypotheses would be worth considering in our system. In A. franciscana, amplexus can last from hours to days and females are found in amplexus even when carrying fertilized eggs (Belk 1991). Mate guarding has been shown to be costly for females in many species (see Arnqvist and Rowe 2005b; Jormalainen 1998 for reviews). These costs (e.g., decreased foraging activity) are possibly stronger if contemporary males guard their mates more than their counterparts from the past or future. Hence, one explanation for the fluctuating selection pattern would be that in our experiment contemporary males were on average either faster to clasp their mates and/or better able to secure her. Under this hypothesis, we expect contemporary males to perform better when competing with noncontemporary males. Measuring the fitness advantage of contemporary males over noncontemporary males when competing for fertilization would help confirm that sexual conflicts are driving male–female coevolution in this species.

TIME-SHIFT EXPERIMENTS AND COEVOLUTION PATTERNS

Time-shift experiments are powerful tools to investigate the dynamics of adaptation to changing selection (e.g., Hairston et al. 1999), especially antagonistic coevolutions (Gaba and Ebert 2009). Validity of inferences on ancient communities based on dormant eggs using this approach is a major concern for resurrection ecologists (e.g., Jankowski and Straile 2003; Keller and Spaak 2004). Such studies usually rely on the hypothesis that dormant eggs represent an unbiased sample of the study population regarding the focal trait. Hatching success of ancient eggs is usually low (e.g., Amat et al. 2005; Jankowski and Straile 2003), which is likely to impose selection on dormancy traits such as basal metabolic rate or diapause duration. Hence, our results rely on the assumption that even if selective hatching would be correlated to sexually antagonistic traits, it would not create the male–female interaction observed here. Although extremely challenging, a formal test of this assumption would be useful.

One conceptual issue in this time-shift experiment is that the model including ΔT along with female and male year effects cannot be mathematically defined. This constraint limits the interpretation of additive effects of female and male years because they cannot be both estimated independently from ΔT effect. However, based on the model selection presented, it appears that the additive male year factor when fitted with ΔT had a persistently weak effect on female life-history traits.

In the last decades, experimental evolution with microbes (virus, bacteria, yeast or unicellular algae) has become the main approach for studying adaptation over many generations in the laboratory (see Bell 2008). Although this method is very informative, research on individual life histories using these model organisms remains limited. Resurrection ecology has recently flourished with the use of Daphnia as a good multicellular alternative to microbes, especially in the study of host–parasite coevolution (e.g., Decaestecker et al. 2007) and adaptation to a novel environment (e.g., Hairston et al. 1999). As we demonstrated in this study, Artemia long-lived resting eggs, could well become a valuable resurrection model for the study of sexual selection in natural conditions. Likewise angiosperm dormant seeds might prove to be a good system to study parental conflicts over postfertilization offspring resource allocation, because this kind of sexual conflicts is likely to be widespread in this group (Friedman 2001; Cailleau et al. 2010).


Associate Editor: N. Barton

AKNOWLEDGMENTS

We thank G. Van Stappen for providing us with the cysts from Artemia franciscana and W. A. Wurtsbaugh for the information about Artemia population dynamics in the Great Salt Lake. We are grateful to V. Hornsperger, A. Remy, E. Flaven for their help with laboratory work and to the ECOGEV team, G. Martin, L.-M. Chevin, F. Blanquart, and O. Gimenez for their help and comments. We also wish to thank T. Chapman and two anonymous reviewers for their comments. Financial support was provided by an ANR grant to AC (ANR-08-JCJC-0041–01), the QuantEvol ERC grant to TL, and a French Ministry of Research fellowship to NR.

Ancillary