Accepted Article
Received Date : 31-May-2014 Revised Date : 30-Nov-2014 Accepted Date : 14-Jan-2015 Article type
: Research Papers
Selection and evolutionary potential of spring arrival phenology in males and females of a migratory songbird
Authors: Tarka M*, Hansson B, Hasselquist D
*
Correspondence: Maja Tarka, Centre for Biodiversity Dynamics, Norwegian University of Science and
Technology, Realfagsbygget, NTNU, 7491 Trondheim, Norway. Phone: +47-73596045. E-mail:
[email protected]
Bengt Hansson, Molecular Ecology and Evolution Lab, Department of Biology, Ecology Building, Sölvegatan 37, Lund University, SE-223 62 Lund, Sweden. Phone: +46-46-2224996; Fax: +46-46-2224716. E-mail: [email protected] Dennis Hasselquist, Molecular Ecology and Evolution Lab, Department of Biology, Ecology Building, Sölvegatan 37, Lund University, SE-223 62 Lund, Sweden. Phone: +46-46-2223708; Fax: +46-46-2224716. E-mail: [email protected]
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/jeb.12638 This article is protected by copyright. All rights reserved.
Accepted Article
Running title: Quantitative genetics of arrival date
Abstract The timing of annual life-history events impacts survival and reproduction of all organisms. A changing environment can perturb phenological adaptations and an important question is if populations can evolve fast enough to track the environmental changes. Yet, little is known about selection and evolutionary potential of traits determining the timing of crucial annual events. Migratory species, which travel between different climatic regions, are particularly affected by global environmental changes. To increase our understanding of evolutionary potential and selection in timing traits, we investigated the quantitative genetics of arrival date at the breeding ground using a multigenerational pedigree of a natural great reed warbler (Acrocephalus arundinaceus) population. We found significant heritability of 16.4% for arrival date and directional selection for earlier arrival in both sexes acting through reproductive success, but not through lifespan. Mean arrival date advanced with 6 days over 20 years, which is in exact accordance with our predicted evolutionary response based on breeder’s equation. However, this phenotypic change is unlikely to be caused by microevolution, because selection seems mainly to act on the non-genetic component of the trait. Furthermore, demographic changes could also not account for the advancing arrival date. Instead, a strong correlation between spring temperatures and population mean arrival date suggests that phenotypic plasticity best explains the advancement of arrival date in our study population. Our study dissects the evolutionary and environmental forces that shape timing traits and thereby increases knowledge of how populations cope with rapidly changing environments. Key words: Migration, phenology, adaptation, microevolution, fitness, Acrocephalus arundinaceus
This article is protected by copyright. All rights reserved.
Accepted Article
Introduction The timing of activities over the annual cycle has important consequences for fitness in natural populations. One crucial yearly activity is the arrival or appearance at the breeding site after migration, dormancy or hibernation. Early arrival at the breeding site is associated with higher reproductive success in a wide range of species, including in fish (Dickerson et al. 2005), insects (Fagerström & Wiklund 1982) and birds (e.g. Aebischer et al. 1996; Lozano et al. 1996). In birds, advantages for early arriving individuals include occupation of better territories and breeding sites, more or higher quality mates and a longer period available for breeding allowing for replacement broods and additional matings (Hasselquist 1998; Newton 2008; Johansson & Jonzén 2012). However, arriving early often carries costs, such as riskier migration and lower food availability, and it is therefore often found that early arriving birds are of higher quality or condition (Alerstam & Lindström 1990; Møller 1994; Kokko 1999; Forstmeier 2002).
The importance of arrival date as a life-history trait strongly connected to fitness, both directly
and indirectly, should be particularly pronounced in long-distance migrant birds, where early spring arrival is essential when coping with the relatively shorter breeding period and the need for an early, yet well-prepared, autumn migration (Alerstam 1991a; Hemborg et al. 2001; Newton 2008). However, the mechanisms of timing of arrival are largely unknown for long-distance migrants, because the condition of the arriving individuals is difficult to predict as they depart from a different climate zone at the onset of their migration (Newton 2008). Suggested important cues for onset of migration are day length and endogenous clocks (Berthold 1996; Gwinner 1996), i.e. cues that do not shift with climate changes. Hence, there is an increasing concern for how migrating species will cope with the current, rapid environmental changes and how increasing spring temperatures will affect the annual timing programs
This article is protected by copyright. All rights reserved.
Accepted Article
of plants and animals (Wilcove & Wikelski 2008). Several phenology studies show that many bird species arrive earlier to their breeding grounds over the course of the last decades, which coincides with increasingly warmer spring temperatures (Cotton 2003; Jonzén et al. 2006; Pulido 2007; Rubolini et al. 2007). A problem arises when the rate of shift in phenology differs between trophic levels, e.g. insects respond faster or with larger timing shifts than birds. This results in a mismatch between the food source peak and the timing of breeding in insectivorous birds causing decreased reproductive success (Both & Visser 2001; Visser 2008) and even population decline (Both et al. 2006, but see Reed et al.
2013). To understand to which extent migratory birds can adapt to the novel selection regimes posed by
changing environments, we need to investigate the genetic basis, and the strength and pattern of selection acting on arrival date. Arrival date is a complex behavioural trait, affected by physiology, morphology and by environmental factors such as food availability, weather and wind (Alerstam & Lindström 1990; Gordo 2007; Pulido 2007). Therefore, the genetic component is likely small in relation to the environmental influence. Nevertheless, the limited data that are available report on heritability (h2) of arrival date ranging from zero to c. 50% depending on species and estimation method (Rees 1989; Potti 1998; Møller 2001; Teplitsky et al. 2011; Arnaud et al. 2012). These studies represent the only data presently available on the genetic basis of arrival date in migrant birds, and hence many of the central aspects of the evolvability of this important fitness trait are still unexplored. Moreover, most studies have solely evaluated directional selection on arrival date in males, and have thus neglected the patterns in females as well as any potential nonlinear effects (but see Møller 1994; Smith & Moore 2005; Møller et al. 2009; Arnaud et al. 2012). If conflicting selection patterns operate in females, this could constrain a response to directional selection in males (e.g. Rice & Chippindale 2001; Tarka et al. 2014). There are several possible biological reasons for the existence of stabilizing selection on arrival date. Too early arrival might be disfavoured because (i) rapid migration might be energetically and
This article is protected by copyright. All rights reserved.
Accepted Article
physiologically costly and lead to poor condition at arrival (Alerstam 1991b), (ii) the environment at the breeding ground might be harsh early in the season (e.g. Brown & Brown 2000; Jonzén et al. 2007)
resulting in low food availability and a decrease in body condition, and (iii) the territory quality might be difficult to judge very early in the season due to lack of reliable cues (Tye 1992; Hansson et al. 2000a). Arriving too late may also incur costs because (a) there may be no or only low quality territories available (Bensch & Hasselquist 1991; Hasselquist & Bensch 1991), (b) decreasing prospects of finding unpaired/fertilizable mates (Bensch & Hasselquist 1992), (c) the food peak might be missed at the time of maximum nestling feeding (Visser et al. 1998; Thomas et al. 2001) , (d) less time for re-nesting if exposed to nest predation (Hansson et al. 2000a), and (e) limited time to prepare for autumn migration both for parents and offspring (Bensch et al. 1985; Hemborg et al. 2001). Furthermore, many species have a protandrous system where males arrive before the females (Fagerström & Wiklund 1982), with potentially very interesting sex-specific selection pressures (Morbey & Ydenberg 2001). Therefore, to understand the evolutionary potential and selection acting on arrival date we need to study the trait in both sexes, as well as investigate both linear and non-linear selection regimes.
Moreover, to get a clear picture it is important to evaluate the relative impact of genes and
environment on phenotypic changes. If both fitness and the trait of interest are associated with condition, an indirect, environmentally driven, covariance between fitness and trait may arise. If we are to understand the evolutionary response of arrival to climate change, this environmental covariance has to be distinguished from the genetic covariance. This is because only a genetic covariance between trait and fitness will lead to an evolutionary response (Robertson 1966; Price 1970; Price et al. 1988). In general, selection estimated on the phenotypic level does not always reflect selection acting on the genetic level, and this can affect the conclusions about predicted evolutionary responses (Merilä et al.
This article is protected by copyright. All rights reserved.
Accepted Article
2001; Morrissey et al. 2012). In a meta-analysis, Gienapp et al. (2007) used an indirect method to disentangle the importance of the genetic and plastic responses causing the advancement in arrival date found in several bird species. One conclusion from their study was that detailed, individually based, long-term studies would be key to distinguish between microevolutionary and environmentally plastic changes, but that no such studies had yet been conducted.
In the present study, we conduct a comprehensive quantitative genetic analysis of the potential
for evolutionary change of arrival date in a migratory songbird by investigating the genetic basis and patterns of selection acting on arrival date. To achieve this goal, we use long-term data from a detailed 27 years long study of a natural population of great reed warblers (Acrocephalus arundinaceus) in
southern Central Sweden. The great reed warbler is a long-distance migrant that spends the winter in tropical Africa (Cramp & Brooks 1992; Lemke et al. 2013). It is a protandrous and facultatively polygynous species, where early-arriving males attract more females to their territories than males that arrive later in the season (Hasselquist 1998). We investigate both directional and quadratic standardized selection acting on arrival date for three fitness components (lifespan, lifetime number of fledglings and lifetime number of recruits) in both male and female great reed warblers. We further evaluate the repeatability of arrival date between years within individuals, and the heritability of arrival date using a multigenerational pedigree in an ‘animal model’ approach (Kruuk 2004; Wilson et al. 2010). Furthermore, we investigate if arrival date has changed over time in the population and how this change correlates with predictions of evolutionary response to selection, demographic patterns and environmental fluctuations between years. Finally, we attempt to discriminate between selection acting on environmental versus genetic components, to investigate the potential for microevolution of spring arrival date and hence adaptation to a changing environment.
This article is protected by copyright. All rights reserved.
Accepted Article
Methods Study population and data collection The great reed warbler breeds in dense reed beds of eutrophic lakes in parts of Europe and Asia. It is a long-distance migrant songbird that conducts solitary, mainly nocturnal, flights to overwintering sites in tropical Africa (De Roo & Deheegher 1969; Cramp & Brooks 1992, Lemke et al. 2013). Our study population breeds at Lake Kvismaren (59°10’N, 15°24’E), in southern Central Sweden. This study population has been monitored on an individual basis since 1983, using individually unique aluminium and colour ring combinations on the birds’ legs (Bensch et al. 1998; Hansson et al. 2004; Åkesson et al. 2007a). Great reed warblers show strong site fidelity and we know from detailed studies of the species’ whole distribution in Sweden that between-year breeding dispersal is very low in this population (Hansson et al. 2002). We estimate that among the birds that survive the winter, ca. 50% of the nestlings and 92% of the adults return to the same lake as the previous year (Bensch et al. 1998; Hansson et al. 2000a, 2002). Mean lifespan of birds that survive their first year is approximately 2.5 years and the oldest individuals (one male and one female) recorded in the population were 9 years old. Population size has varied from 18 to 78 adult individuals in different years with an average of 52 adult birds per year.
Data on arrival date used in this study have been collected in the years 1985-2004. For the core arrival period spanning early May to early July, each potential territory was visited virtually on a daily basis (every 1-3 day). Our daily field observations strongly suggest that males in most cases can occupy a territory and start to sing to attract a female within 0-3 days after arrival (Bensch & Hasselquist 1992; cf. Romero et al. 1997; Fransson & Jakobsson 1998). We also know from geolocator studies that the arrival date estimated from observations corresponds well with actual arrival date of the birds (Lemke et al.
This article is protected by copyright. All rights reserved.
Accepted Article
2013; Tarka M., Hansson B. and Hasselquist D., unpublished data). Hence, for the males, arrival date was recorded as the first day a male was singing in the breeding area (Bensch et al. 1998). Female great reed warblers do not sing, but male singing behaviour changes drastically (from ‘long song’ to ‘short song’) immediately after a female settles in his territory (Catchpole 1983; Hasselquist & Bensch 1991; Bensch et al. 1998). The male continues singing short song when the female is nest building (4-6 days), but around the time the female lays her first egg, he resumes singing long song allowing him the possibility to attract another female (Hasselquist & Bensch 1991). We also know from radio-track, release experiments that females unfamiliar to our study area choose a male and his territory within 1-3 days after arrival (Bensch & Hasselquist 1992). Female arrival date was hence recorded as the first date she settled in a territory. Based on our knowledge of this species, and because all individuals have been treated in the same way, we believe that the estimated arrival day is comparable between individuals. However, there is a risk that some of the late arriving birds did not settle in Lake Kvismaren directly after their spring migration, but instead had arrived earlier at another site where they may have even tried to breed. It is difficult to distinguish such site-changing individuals from ‘truly’ late-arriving birds. Arrival date is negatively associated with age (old birds arrive earlier than young birds) and more than 90% of the birds arriving late in the season were first year birds that have commenced their first spring migration and may therefore be less efficient migrants (Tarka M., Hansson B. and Hasselquist D., unpublished data). Hence, we cannot determine a reliable cut-off date to separate site-changing individuals from true late-comers, and therefore to be conservative we included all data points. In total, we have recorded 1035 arrivals, based on 548 unique individuals (295 females and 253 males). Arrival date was recorded as consecutive numbers, where 1 = 1st of May. To obtain pedigree data, we have used both field observations (1983-2004) and molecular
methods (1987-2004) to confirm true biological parents of young for each brood. Extra-pair paternity is rare, only 3% of the young in this population have been sired by another male than their social ‘father’
This article is protected by copyright. All rights reserved.
Accepted Article
(Hasselquist et al. 1995, 1996; Arlt et al. 2004; Hansson et al. 2006), and these cases have been corrected in the pedigree. This work has led to an extensive pedigree with nearly all relationships in the population resolved (Hansson et al. 2005; Åkesson et al. 2007b). When including only individuals with records on arrival date and informative links with relatives (Morrissey & Wilson 2010), our pedigree contains 550 records and a maximum pedigree depth of 7 generations. This includes 210 maternities, 214 paternities, 92 full sibs, 331 half sibs, 80 maternal grandmothers, 80 maternal grandfathers, 99 paternal grandmothers and 102 paternal grandfathers (R package pedantics; Morrissey & Wilson 2010).
Fitness estimates and selection We used three different fitness components: (i) lifetime fledgling success (LFS), which is the total number of fledglings produced by an individual in its lifetime, (ii) lifetime recruit success (LRS), which is the total number of recruited offspring to the population (i.e., offspring returning to breed when ≥ 1
year old) in an individual’s lifetime, and (iii) lifespan, which is the age of the individual the last time it was recorded in the population. These estimates are not completely independent of each other, but still provide us with somewhat different information. Both LFS and LRS include lifespan, because they are a sum of the reproductive output for all years an individual was present in the population. LFS is more normally distributed, and less biased by potentially stochastic events, such as offspring deaths occurring during the first migration or dispersal to other breeding sites, than LRS. However, LRS provides better information than LFS about which individuals pass on genes to the local population in subsequent generations of breeders. LRS has a skewed distribution with many zeros, because the majority of individuals do not produce any recruits. Lifespan differs from LFS and LRS because it does not include reproductive output, instead reflecting survival and longevity. This is an important fitness component to analyse, because there might be a potential trade-off between migratory efforts (and hence arrival date)
This article is protected by copyright. All rights reserved.
Accepted Article
and longevity. Lifespan could be biased by confounding factors such as dispersal, but we know that between-year breeding dispersal is very low in this population (92% return to breed at their former breeding site; Hansson et al. 2002). Data on fitness were collected until 2010 to obtain lifetime fitness for all individuals from the cohorts hatched in 1984-2003, measured on arrival date in the years 19852004.
To obtain standardized selection differentials, we used standard regression methods (Lande & Arnold 1983; Arnold & Wade 1984; Brodie III et al. 1995). Fitness was divided by the overall population mean to obtain relative fitness separately for each model. Because the dependent variable (the lifetime fitness component) is a single measurement per individual, we used age- and year-corrected (and in pooled analyses also sex-corrected to account for protandry) individual means of arrival date (the independent variable). Using the individual means has the disadvantage of excluding the within-individual variance. However, we wanted to use lifetime reproductive success and lifespan as our fitness-related measures, and hence needed a single independent variable per individual, instead of annual reproductive success which would be the option for multiple measurements. To investigate if our estimates were affected by using individual means (see e.g., Taylor et al. 2014), we also ran a model based on annual reproductive success and multiple measurements of arrival date for each individual. We found that the estimates and standard errors from the individual mean models were very similar to the repeated measures models (see Results). Hence we believe that our selection estimates based on individual means are conservative, and we only present comprehensive results from these models. Arrival date is dependent on age, where older individuals arrive earlier than young ones up till 3
years of age (Bensch 1996; Hasselquist et al. 1996), and we therefore divided age into three classes: 1year old, 2-year old and ≥3-year old individuals. Arrival date varies between years due to environmental
This article is protected by copyright. All rights reserved.
Accepted Article
factors such as weather conditions; thus, we also corrected for year. The correction was done by adjusting each measurement with the estimated effect sizes of age and year (and in the case of models with pooled sexes also sex), which were obtained from general linear mixed models with individual ID as a random factor. This adjusted all the measurements to the expected arrival of 1-year-old individuals arriving during the first year of study. The individual means of corrected arrival dates were further standardized by subtracting the mean and dividing by the standard deviation to make the estimates comparable between studies (Brodie III et al. 1995). This standardization was done separately for each model (males, females and sexes pooled). Directional selection estimates were obtained from a general linear model, regressing relative fitness on arrival date, and estimates for quadratic selection were taken from a linear model including both arrival date and squared arrival date (Lande & Arnold 1983; Arnold & Wade 1984). The estimate and its standard error for quadratic selection was doubled to be comparable with directional selection (Stinchcombe et al. 2008). Because of a non-normal distribution of all three fitness estimates, the statistical significance was determined by bootstrapping the 95% confidence interval for the regression coefficients using 10000 replicates (Brodie III et al. 1995; R package boot; Canty & Ripley 2014). Selection was visualized using cubic splines (R package mgcv; Wood 2011).
Repeatability and heritability An ‘animal model’ approach was used to estimate repeatability, genetic components and heritability. The ‘animal model’ is a general linear mixed model that simultaneously estimates fixed effects and divides the remaining trait variance into genetic and environmental components by utilizing the pedigree links (Kruuk 2004; Wilson et al. 2010). To adjust to a normal distribution, we square-root
transformed arrival date before fitting the model (for this analysis we shifted the data with 5 days to avoid negative values due to a few individuals arriving before 1st of May). We used age (same three
This article is protected by copyright. All rights reserved.
Accepted Article
classes as in the selection analysis), and in the case of pooled data also sex, as fixed effects. As random effects, we fitted the additive genetic effect (individual ID connected to the pedigree; ‘animal’ in Equation 1), permanent environment effect (individual ID not connected to the pedigree to account for permanent between-individual variation and repeated measures; ‘ID’ in Equation 1), year effects (to account for seasonal variations; ‘Year2’ in Equation 1) and maternal effects (mother ID to account for non-genetic effects of mother and nest; ‘Mother’ in Equation 1). Additionally, we also tested a model including year as a fixed covariate (‘Year1’ in Equation 1) to account for possible trends over years. The
final model was as follows: sqrt(Arrival date) = Age + Year1 + Sex + animal + ID + Year2 + Mother.
(Equation 1)
To investigate if the additive genetic variance differed between the sexes and to estimate the strength of intersexual genetic correlation, we used a bivariate animal model with male and female arrival date as response variables. Sex was excluded as a fixed factor in this model, and covariances for all random effects, except additive genetic effects, were constrained to zero (Wilson et al. 2010). The models were
run using the software AsReml v.3.0 (Gilmour et al. 2009). Statistical significance of effects was obtained
using log-likelihood ratio tests with 1 degree of freedom from a χ2 distribution (Wilson et al. 2010).
Change in arrival date over time – demography, microevolution or phenotypic plasticity? To test if there has been a change in arrival date during the course of the study (1985-2004), we calculated mean unstandardized arrival date per year and made a linear regression over the 20 years of study. To test for possible biases due to e.g. age- and sex-related demographic patterns that could affect changes in arrival date over time, we used all individual records of arrival date in a linear mixed model with repeated measures (individual ID as random effect), where we tested year, sex, age and all
This article is protected by copyright. All rights reserved.
Accepted Article
interactions as fixed effects. We used maximum likelihood as estimation method and compared nested models using log-likelihood ratio tests (Zuur et al. 2009; Bolker et al. 2009). The R package nlme was used (Pinheiro et al. 2015). The estimates were taken from a model with arrival date not square-root transformed for a better understanding of the effect, and the t- and p-values from a model where arrival date was square-root transformed to obtain a better fit to a normal distribution. To predict microevolutionary change over time we used the breeder’s equation (Lush 1937). For
the breeder’s equation (R=h2*S, where the evolutionary response per generation, R, equals the
heritability, h2, times the selection differential S), we fitted the selection differential based on LRS (recalculated to day units) and genetic variance standardized by the phenotypic variance (i.e. heritability, h2) because arrival date is on an interval scale (Houle et al. 2011; Hansen et al. 2011). To visualize the expected microevolutionary change based on the breeder’s equation in relation to the phenotypic change and shift in spring warming onset, we calculated the cumulative expected change per year (dividing the per generation change with the generation time) with 1985’s mean bird arrival date as the starting phenotype. We note that there are several potential problems with the prediction of microevolution using the breeder’s equation, including the problem to infer the causality of the correlation between trait and fitness in unmanipulated wild populations (Morrissey et al. 2010). However, in many cases, this is the best prediction that can be done.
To complement the prediction from the breeder’s equation, we compared it with the prediction
from the Robertson-Price identity (Morrissey et al. 2010). The selection differential can sometimes be driven by environmental factors, such as condition, that affect both the phenotype and fitness and create a non-genetic correlation (cf. Price et al. 1988b; Merilä et al. 2001). By using the Robertson-Price Identity this potential problem can be avoided and selection acting directly on the genetic component of
This article is protected by copyright. All rights reserved.
Accepted Article
a trait is estimated (Morrissey et al. 2010). To distinguish if selection was acting on the genetic or the environmental component of arrival date, we fitted a bivariate ‘animal model’ with relative fitness and
individual mean arrival date as response variables. Here, we used the same data set as for the selection estimates (corrected (age, year) individual mean arrival date (here not transformed to mean 0 and SD 1) and relative lifetime recruit success calculated as above). The fitted model included sex as a fixed factor and additive genetic effect as a random factor. We used the software AsReml v.3.0 (Gilmour et al. 2009)
and significance of correlations was obtained by using log-likelihood ratio test between a model with the covariance varying freely and a model where the covariance was constrained to zero (Wilson et al. 2010).
Fine-grained population responses to environmental changes can be used to infer phenotypic plasticity as a likely explanation for phenotypic changes over time. If the population mean strongly correlates with an environmental trait on a year-to-year basis, phenotypic plasticity becomes a likely explanation (Merilä & Hendry 2014). To investigate how population mean arrival date correlates with yearly fluctuations in the local environment at the breeding site, we calculated the yearly onset of spring warming in Kvismaren. To obtain a standardized timing of the start of the spring, i.e. onset of spring warming, at the breeding site each year, we used a similar approach to that of (Saino et al. 2011) based on a calculation of accumulating winter and spring temperatures. Each year, when the cumulative sum of the daily mean temperatures reached above a set level (see below), the date when this occurred was used as the date for the onset of spring warming. Daily mean temperatures for the years 1980-2004 were downloaded from the Swedish Meteorological and Hydrological Institute (www.SMHI.se; assessed September 2013). We used data from three weather stations in the area around our study site (Örebro: 59°28’N, 15°16’E; Karlstad Flygplats: 59°44’N, 13°34’E; Jönköpings Flygplats: 57°75’N, 14°07’E) and
This article is protected by copyright. All rights reserved.
Accepted Article
calculated the temperature means for each day based on these. Then we calculated the accumulation of ‘degree-days’ (DD) with a threshold 0°C for the instantaneous temperatures (Saino et al. 2011) and the spring ‘mean arrival date’ (MAD) based on raw data of the yearly averages of the years 1985-1987. This MAD was then used to calculate DD for the years 1980-1984 and the average DD of these years was then used as a baseline (DDbase) for the spring-onset defining temperatures. For each of the study years 1985-2004, we cumulatively summed the temperatures (above 0°C) from 1st of January until DDbase was reached. The date when the temperature passed the DDbase threshold was used as the date of the onset of spring warming for that particular year.
Results Mean uncorrected arrival date for males was 19th of May (SD = 14 days, N=253) and for females 30th of May (SD = 12 days, N=295). Mean arrival date of each year’s first male was 3rd of May (SD = 3 days) and
for the first female it was 14th of May (SD = 5 days).
Selection on arrival date We found negative directional selection on arrival date for the sexes combined, as well as for males and females separately, both when using lifetime fledgling success (LFS) and lifetime recruiting offspring success (LRS) as fitness estimates (estimates of standardized selection differentials ranging from -0.226
± 0.058 to -0.368 ± 0.099; Table 1). This means that in both sexes, individuals that arrived earlier in
spring produced relatively more fledglings and recruiting offspring over their lives than the population average. The estimates and standard errors from these individual mean models (sexes combined; LFS: 0.267 ± 0.052; LRS: -0.341 ± 0.071; see Table 1) were very similar to estimates from models using
This article is protected by copyright. All rights reserved.
Accepted Article
repeated measures (sexes combined; Annual FS: -0.280 ± 0.033; Annual RS: -0.332 ± 0.065). For lifespan, the directional selection estimates on arrival date were lower (ranging from -0.021 to -0.051) and non-significant (Table 1). When analysing quadratic selection on arrival date with respect to lifespan, there was a
significant stabilizing selection in females and for the sexes combined. In males, the estimate of stabilizing selection was of the same magnitude as in females; however it was not significant (Table 1). One problem with using individual means to estimate quadratic selection is that individuals that live longer will be repeatedly measured and therefore their mean values are expected to vary less compared to short lived individuals that only occur for one or a few years in the population. This unequal variance between short- and long-lived birds could potentially generate a pattern between arrival date and lifespan that resembles stabilizing selection (individuals with high lifespan have values with low variance, and individuals with short lifespan have values with high variance, around the mean standardized arrival date, and when a curve is fitted to such data it will include a negative quadratic component). To avoid this potentially spurious effect, we reran our arrival date–lifespan analysis including only (year and age standardized) arrival date from one random year for each individual. This analysis showed that in all three cases (males, females and sexes combined) the quadratic estimates of lifespan selection were lower in magnitude compared with the analysis based on means, and also not statistically significant (quadratic selection differential estimates (doubled) with 95% CI [lower, upper]; males: 0.024 [-0.072, 0.114]; females: -0.038 [-0.114, 0.42]; total: -0.010 [-0.070, 0.050]). This suggests that the significant estimate of quadratic selection acting through lifespan might be inflated and therefore should be treated with caution. Cubic splines that visualise the selection acting on arrival date (Figure 1a-c) largely confirmed
the patterns found in the analyses of selection differentials (Table 1). For LFS (Figure 1a) and LRS (Figure
This article is protected by copyright. All rights reserved.
Accepted Article
1b), the curve had a negative slope with a peak left of the population mean, which indicates that there is pronounced negative directional selection acting on arrival date through these two fitness estimates. Regarding lifespan (Figure 1c), the curve was more hump-shaped and the optimum is closer to the population mean, which is in line with the analyses of selection differentials where we found weak nonsignificant directional selection and some evidence in support of stabilizing selection (Table 1, but see the caveats above).
Repeatability and heritability of arrival date Repeatability of arrival date was moderate and significant (0.366 ± 0.040 SE, P < 0.001), meaning that 36.6% of the phenotypic variance is explained by individuals having consistent arrival date between years. Furthermore, we found a significant heritability of arrival date of 16.4% (h2 = 0.164 ± 0.065 SE, P
Received Date : 31-May-2014 Revised Date : 30-Nov-2014 Accepted Date : 14-Jan-2015 Article type
: Research Papers
Selection and evolutionary potential of spring arrival phenology in males and females of a migratory songbird
Authors: Tarka M*, Hansson B, Hasselquist D
*
Correspondence: Maja Tarka, Centre for Biodiversity Dynamics, Norwegian University of Science and
Technology, Realfagsbygget, NTNU, 7491 Trondheim, Norway. Phone: +47-73596045. E-mail:
[email protected]
Bengt Hansson, Molecular Ecology and Evolution Lab, Department of Biology, Ecology Building, Sölvegatan 37, Lund University, SE-223 62 Lund, Sweden. Phone: +46-46-2224996; Fax: +46-46-2224716. E-mail: [email protected] Dennis Hasselquist, Molecular Ecology and Evolution Lab, Department of Biology, Ecology Building, Sölvegatan 37, Lund University, SE-223 62 Lund, Sweden. Phone: +46-46-2223708; Fax: +46-46-2224716. E-mail: [email protected]
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/jeb.12638 This article is protected by copyright. All rights reserved.
Accepted Article
Running title: Quantitative genetics of arrival date
Abstract The timing of annual life-history events impacts survival and reproduction of all organisms. A changing environment can perturb phenological adaptations and an important question is if populations can evolve fast enough to track the environmental changes. Yet, little is known about selection and evolutionary potential of traits determining the timing of crucial annual events. Migratory species, which travel between different climatic regions, are particularly affected by global environmental changes. To increase our understanding of evolutionary potential and selection in timing traits, we investigated the quantitative genetics of arrival date at the breeding ground using a multigenerational pedigree of a natural great reed warbler (Acrocephalus arundinaceus) population. We found significant heritability of 16.4% for arrival date and directional selection for earlier arrival in both sexes acting through reproductive success, but not through lifespan. Mean arrival date advanced with 6 days over 20 years, which is in exact accordance with our predicted evolutionary response based on breeder’s equation. However, this phenotypic change is unlikely to be caused by microevolution, because selection seems mainly to act on the non-genetic component of the trait. Furthermore, demographic changes could also not account for the advancing arrival date. Instead, a strong correlation between spring temperatures and population mean arrival date suggests that phenotypic plasticity best explains the advancement of arrival date in our study population. Our study dissects the evolutionary and environmental forces that shape timing traits and thereby increases knowledge of how populations cope with rapidly changing environments. Key words: Migration, phenology, adaptation, microevolution, fitness, Acrocephalus arundinaceus
This article is protected by copyright. All rights reserved.
Accepted Article
Introduction The timing of activities over the annual cycle has important consequences for fitness in natural populations. One crucial yearly activity is the arrival or appearance at the breeding site after migration, dormancy or hibernation. Early arrival at the breeding site is associated with higher reproductive success in a wide range of species, including in fish (Dickerson et al. 2005), insects (Fagerström & Wiklund 1982) and birds (e.g. Aebischer et al. 1996; Lozano et al. 1996). In birds, advantages for early arriving individuals include occupation of better territories and breeding sites, more or higher quality mates and a longer period available for breeding allowing for replacement broods and additional matings (Hasselquist 1998; Newton 2008; Johansson & Jonzén 2012). However, arriving early often carries costs, such as riskier migration and lower food availability, and it is therefore often found that early arriving birds are of higher quality or condition (Alerstam & Lindström 1990; Møller 1994; Kokko 1999; Forstmeier 2002).
The importance of arrival date as a life-history trait strongly connected to fitness, both directly
and indirectly, should be particularly pronounced in long-distance migrant birds, where early spring arrival is essential when coping with the relatively shorter breeding period and the need for an early, yet well-prepared, autumn migration (Alerstam 1991a; Hemborg et al. 2001; Newton 2008). However, the mechanisms of timing of arrival are largely unknown for long-distance migrants, because the condition of the arriving individuals is difficult to predict as they depart from a different climate zone at the onset of their migration (Newton 2008). Suggested important cues for onset of migration are day length and endogenous clocks (Berthold 1996; Gwinner 1996), i.e. cues that do not shift with climate changes. Hence, there is an increasing concern for how migrating species will cope with the current, rapid environmental changes and how increasing spring temperatures will affect the annual timing programs
This article is protected by copyright. All rights reserved.
Accepted Article
of plants and animals (Wilcove & Wikelski 2008). Several phenology studies show that many bird species arrive earlier to their breeding grounds over the course of the last decades, which coincides with increasingly warmer spring temperatures (Cotton 2003; Jonzén et al. 2006; Pulido 2007; Rubolini et al. 2007). A problem arises when the rate of shift in phenology differs between trophic levels, e.g. insects respond faster or with larger timing shifts than birds. This results in a mismatch between the food source peak and the timing of breeding in insectivorous birds causing decreased reproductive success (Both & Visser 2001; Visser 2008) and even population decline (Both et al. 2006, but see Reed et al.
2013). To understand to which extent migratory birds can adapt to the novel selection regimes posed by
changing environments, we need to investigate the genetic basis, and the strength and pattern of selection acting on arrival date. Arrival date is a complex behavioural trait, affected by physiology, morphology and by environmental factors such as food availability, weather and wind (Alerstam & Lindström 1990; Gordo 2007; Pulido 2007). Therefore, the genetic component is likely small in relation to the environmental influence. Nevertheless, the limited data that are available report on heritability (h2) of arrival date ranging from zero to c. 50% depending on species and estimation method (Rees 1989; Potti 1998; Møller 2001; Teplitsky et al. 2011; Arnaud et al. 2012). These studies represent the only data presently available on the genetic basis of arrival date in migrant birds, and hence many of the central aspects of the evolvability of this important fitness trait are still unexplored. Moreover, most studies have solely evaluated directional selection on arrival date in males, and have thus neglected the patterns in females as well as any potential nonlinear effects (but see Møller 1994; Smith & Moore 2005; Møller et al. 2009; Arnaud et al. 2012). If conflicting selection patterns operate in females, this could constrain a response to directional selection in males (e.g. Rice & Chippindale 2001; Tarka et al. 2014). There are several possible biological reasons for the existence of stabilizing selection on arrival date. Too early arrival might be disfavoured because (i) rapid migration might be energetically and
This article is protected by copyright. All rights reserved.
Accepted Article
physiologically costly and lead to poor condition at arrival (Alerstam 1991b), (ii) the environment at the breeding ground might be harsh early in the season (e.g. Brown & Brown 2000; Jonzén et al. 2007)
resulting in low food availability and a decrease in body condition, and (iii) the territory quality might be difficult to judge very early in the season due to lack of reliable cues (Tye 1992; Hansson et al. 2000a). Arriving too late may also incur costs because (a) there may be no or only low quality territories available (Bensch & Hasselquist 1991; Hasselquist & Bensch 1991), (b) decreasing prospects of finding unpaired/fertilizable mates (Bensch & Hasselquist 1992), (c) the food peak might be missed at the time of maximum nestling feeding (Visser et al. 1998; Thomas et al. 2001) , (d) less time for re-nesting if exposed to nest predation (Hansson et al. 2000a), and (e) limited time to prepare for autumn migration both for parents and offspring (Bensch et al. 1985; Hemborg et al. 2001). Furthermore, many species have a protandrous system where males arrive before the females (Fagerström & Wiklund 1982), with potentially very interesting sex-specific selection pressures (Morbey & Ydenberg 2001). Therefore, to understand the evolutionary potential and selection acting on arrival date we need to study the trait in both sexes, as well as investigate both linear and non-linear selection regimes.
Moreover, to get a clear picture it is important to evaluate the relative impact of genes and
environment on phenotypic changes. If both fitness and the trait of interest are associated with condition, an indirect, environmentally driven, covariance between fitness and trait may arise. If we are to understand the evolutionary response of arrival to climate change, this environmental covariance has to be distinguished from the genetic covariance. This is because only a genetic covariance between trait and fitness will lead to an evolutionary response (Robertson 1966; Price 1970; Price et al. 1988). In general, selection estimated on the phenotypic level does not always reflect selection acting on the genetic level, and this can affect the conclusions about predicted evolutionary responses (Merilä et al.
This article is protected by copyright. All rights reserved.
Accepted Article
2001; Morrissey et al. 2012). In a meta-analysis, Gienapp et al. (2007) used an indirect method to disentangle the importance of the genetic and plastic responses causing the advancement in arrival date found in several bird species. One conclusion from their study was that detailed, individually based, long-term studies would be key to distinguish between microevolutionary and environmentally plastic changes, but that no such studies had yet been conducted.
In the present study, we conduct a comprehensive quantitative genetic analysis of the potential
for evolutionary change of arrival date in a migratory songbird by investigating the genetic basis and patterns of selection acting on arrival date. To achieve this goal, we use long-term data from a detailed 27 years long study of a natural population of great reed warblers (Acrocephalus arundinaceus) in
southern Central Sweden. The great reed warbler is a long-distance migrant that spends the winter in tropical Africa (Cramp & Brooks 1992; Lemke et al. 2013). It is a protandrous and facultatively polygynous species, where early-arriving males attract more females to their territories than males that arrive later in the season (Hasselquist 1998). We investigate both directional and quadratic standardized selection acting on arrival date for three fitness components (lifespan, lifetime number of fledglings and lifetime number of recruits) in both male and female great reed warblers. We further evaluate the repeatability of arrival date between years within individuals, and the heritability of arrival date using a multigenerational pedigree in an ‘animal model’ approach (Kruuk 2004; Wilson et al. 2010). Furthermore, we investigate if arrival date has changed over time in the population and how this change correlates with predictions of evolutionary response to selection, demographic patterns and environmental fluctuations between years. Finally, we attempt to discriminate between selection acting on environmental versus genetic components, to investigate the potential for microevolution of spring arrival date and hence adaptation to a changing environment.
This article is protected by copyright. All rights reserved.
Accepted Article
Methods Study population and data collection The great reed warbler breeds in dense reed beds of eutrophic lakes in parts of Europe and Asia. It is a long-distance migrant songbird that conducts solitary, mainly nocturnal, flights to overwintering sites in tropical Africa (De Roo & Deheegher 1969; Cramp & Brooks 1992, Lemke et al. 2013). Our study population breeds at Lake Kvismaren (59°10’N, 15°24’E), in southern Central Sweden. This study population has been monitored on an individual basis since 1983, using individually unique aluminium and colour ring combinations on the birds’ legs (Bensch et al. 1998; Hansson et al. 2004; Åkesson et al. 2007a). Great reed warblers show strong site fidelity and we know from detailed studies of the species’ whole distribution in Sweden that between-year breeding dispersal is very low in this population (Hansson et al. 2002). We estimate that among the birds that survive the winter, ca. 50% of the nestlings and 92% of the adults return to the same lake as the previous year (Bensch et al. 1998; Hansson et al. 2000a, 2002). Mean lifespan of birds that survive their first year is approximately 2.5 years and the oldest individuals (one male and one female) recorded in the population were 9 years old. Population size has varied from 18 to 78 adult individuals in different years with an average of 52 adult birds per year.
Data on arrival date used in this study have been collected in the years 1985-2004. For the core arrival period spanning early May to early July, each potential territory was visited virtually on a daily basis (every 1-3 day). Our daily field observations strongly suggest that males in most cases can occupy a territory and start to sing to attract a female within 0-3 days after arrival (Bensch & Hasselquist 1992; cf. Romero et al. 1997; Fransson & Jakobsson 1998). We also know from geolocator studies that the arrival date estimated from observations corresponds well with actual arrival date of the birds (Lemke et al.
This article is protected by copyright. All rights reserved.
Accepted Article
2013; Tarka M., Hansson B. and Hasselquist D., unpublished data). Hence, for the males, arrival date was recorded as the first day a male was singing in the breeding area (Bensch et al. 1998). Female great reed warblers do not sing, but male singing behaviour changes drastically (from ‘long song’ to ‘short song’) immediately after a female settles in his territory (Catchpole 1983; Hasselquist & Bensch 1991; Bensch et al. 1998). The male continues singing short song when the female is nest building (4-6 days), but around the time the female lays her first egg, he resumes singing long song allowing him the possibility to attract another female (Hasselquist & Bensch 1991). We also know from radio-track, release experiments that females unfamiliar to our study area choose a male and his territory within 1-3 days after arrival (Bensch & Hasselquist 1992). Female arrival date was hence recorded as the first date she settled in a territory. Based on our knowledge of this species, and because all individuals have been treated in the same way, we believe that the estimated arrival day is comparable between individuals. However, there is a risk that some of the late arriving birds did not settle in Lake Kvismaren directly after their spring migration, but instead had arrived earlier at another site where they may have even tried to breed. It is difficult to distinguish such site-changing individuals from ‘truly’ late-arriving birds. Arrival date is negatively associated with age (old birds arrive earlier than young birds) and more than 90% of the birds arriving late in the season were first year birds that have commenced their first spring migration and may therefore be less efficient migrants (Tarka M., Hansson B. and Hasselquist D., unpublished data). Hence, we cannot determine a reliable cut-off date to separate site-changing individuals from true late-comers, and therefore to be conservative we included all data points. In total, we have recorded 1035 arrivals, based on 548 unique individuals (295 females and 253 males). Arrival date was recorded as consecutive numbers, where 1 = 1st of May. To obtain pedigree data, we have used both field observations (1983-2004) and molecular
methods (1987-2004) to confirm true biological parents of young for each brood. Extra-pair paternity is rare, only 3% of the young in this population have been sired by another male than their social ‘father’
This article is protected by copyright. All rights reserved.
Accepted Article
(Hasselquist et al. 1995, 1996; Arlt et al. 2004; Hansson et al. 2006), and these cases have been corrected in the pedigree. This work has led to an extensive pedigree with nearly all relationships in the population resolved (Hansson et al. 2005; Åkesson et al. 2007b). When including only individuals with records on arrival date and informative links with relatives (Morrissey & Wilson 2010), our pedigree contains 550 records and a maximum pedigree depth of 7 generations. This includes 210 maternities, 214 paternities, 92 full sibs, 331 half sibs, 80 maternal grandmothers, 80 maternal grandfathers, 99 paternal grandmothers and 102 paternal grandfathers (R package pedantics; Morrissey & Wilson 2010).
Fitness estimates and selection We used three different fitness components: (i) lifetime fledgling success (LFS), which is the total number of fledglings produced by an individual in its lifetime, (ii) lifetime recruit success (LRS), which is the total number of recruited offspring to the population (i.e., offspring returning to breed when ≥ 1
year old) in an individual’s lifetime, and (iii) lifespan, which is the age of the individual the last time it was recorded in the population. These estimates are not completely independent of each other, but still provide us with somewhat different information. Both LFS and LRS include lifespan, because they are a sum of the reproductive output for all years an individual was present in the population. LFS is more normally distributed, and less biased by potentially stochastic events, such as offspring deaths occurring during the first migration or dispersal to other breeding sites, than LRS. However, LRS provides better information than LFS about which individuals pass on genes to the local population in subsequent generations of breeders. LRS has a skewed distribution with many zeros, because the majority of individuals do not produce any recruits. Lifespan differs from LFS and LRS because it does not include reproductive output, instead reflecting survival and longevity. This is an important fitness component to analyse, because there might be a potential trade-off between migratory efforts (and hence arrival date)
This article is protected by copyright. All rights reserved.
Accepted Article
and longevity. Lifespan could be biased by confounding factors such as dispersal, but we know that between-year breeding dispersal is very low in this population (92% return to breed at their former breeding site; Hansson et al. 2002). Data on fitness were collected until 2010 to obtain lifetime fitness for all individuals from the cohorts hatched in 1984-2003, measured on arrival date in the years 19852004.
To obtain standardized selection differentials, we used standard regression methods (Lande & Arnold 1983; Arnold & Wade 1984; Brodie III et al. 1995). Fitness was divided by the overall population mean to obtain relative fitness separately for each model. Because the dependent variable (the lifetime fitness component) is a single measurement per individual, we used age- and year-corrected (and in pooled analyses also sex-corrected to account for protandry) individual means of arrival date (the independent variable). Using the individual means has the disadvantage of excluding the within-individual variance. However, we wanted to use lifetime reproductive success and lifespan as our fitness-related measures, and hence needed a single independent variable per individual, instead of annual reproductive success which would be the option for multiple measurements. To investigate if our estimates were affected by using individual means (see e.g., Taylor et al. 2014), we also ran a model based on annual reproductive success and multiple measurements of arrival date for each individual. We found that the estimates and standard errors from the individual mean models were very similar to the repeated measures models (see Results). Hence we believe that our selection estimates based on individual means are conservative, and we only present comprehensive results from these models. Arrival date is dependent on age, where older individuals arrive earlier than young ones up till 3
years of age (Bensch 1996; Hasselquist et al. 1996), and we therefore divided age into three classes: 1year old, 2-year old and ≥3-year old individuals. Arrival date varies between years due to environmental
This article is protected by copyright. All rights reserved.
Accepted Article
factors such as weather conditions; thus, we also corrected for year. The correction was done by adjusting each measurement with the estimated effect sizes of age and year (and in the case of models with pooled sexes also sex), which were obtained from general linear mixed models with individual ID as a random factor. This adjusted all the measurements to the expected arrival of 1-year-old individuals arriving during the first year of study. The individual means of corrected arrival dates were further standardized by subtracting the mean and dividing by the standard deviation to make the estimates comparable between studies (Brodie III et al. 1995). This standardization was done separately for each model (males, females and sexes pooled). Directional selection estimates were obtained from a general linear model, regressing relative fitness on arrival date, and estimates for quadratic selection were taken from a linear model including both arrival date and squared arrival date (Lande & Arnold 1983; Arnold & Wade 1984). The estimate and its standard error for quadratic selection was doubled to be comparable with directional selection (Stinchcombe et al. 2008). Because of a non-normal distribution of all three fitness estimates, the statistical significance was determined by bootstrapping the 95% confidence interval for the regression coefficients using 10000 replicates (Brodie III et al. 1995; R package boot; Canty & Ripley 2014). Selection was visualized using cubic splines (R package mgcv; Wood 2011).
Repeatability and heritability An ‘animal model’ approach was used to estimate repeatability, genetic components and heritability. The ‘animal model’ is a general linear mixed model that simultaneously estimates fixed effects and divides the remaining trait variance into genetic and environmental components by utilizing the pedigree links (Kruuk 2004; Wilson et al. 2010). To adjust to a normal distribution, we square-root
transformed arrival date before fitting the model (for this analysis we shifted the data with 5 days to avoid negative values due to a few individuals arriving before 1st of May). We used age (same three
This article is protected by copyright. All rights reserved.
Accepted Article
classes as in the selection analysis), and in the case of pooled data also sex, as fixed effects. As random effects, we fitted the additive genetic effect (individual ID connected to the pedigree; ‘animal’ in Equation 1), permanent environment effect (individual ID not connected to the pedigree to account for permanent between-individual variation and repeated measures; ‘ID’ in Equation 1), year effects (to account for seasonal variations; ‘Year2’ in Equation 1) and maternal effects (mother ID to account for non-genetic effects of mother and nest; ‘Mother’ in Equation 1). Additionally, we also tested a model including year as a fixed covariate (‘Year1’ in Equation 1) to account for possible trends over years. The
final model was as follows: sqrt(Arrival date) = Age + Year1 + Sex + animal + ID + Year2 + Mother.
(Equation 1)
To investigate if the additive genetic variance differed between the sexes and to estimate the strength of intersexual genetic correlation, we used a bivariate animal model with male and female arrival date as response variables. Sex was excluded as a fixed factor in this model, and covariances for all random effects, except additive genetic effects, were constrained to zero (Wilson et al. 2010). The models were
run using the software AsReml v.3.0 (Gilmour et al. 2009). Statistical significance of effects was obtained
using log-likelihood ratio tests with 1 degree of freedom from a χ2 distribution (Wilson et al. 2010).
Change in arrival date over time – demography, microevolution or phenotypic plasticity? To test if there has been a change in arrival date during the course of the study (1985-2004), we calculated mean unstandardized arrival date per year and made a linear regression over the 20 years of study. To test for possible biases due to e.g. age- and sex-related demographic patterns that could affect changes in arrival date over time, we used all individual records of arrival date in a linear mixed model with repeated measures (individual ID as random effect), where we tested year, sex, age and all
This article is protected by copyright. All rights reserved.
Accepted Article
interactions as fixed effects. We used maximum likelihood as estimation method and compared nested models using log-likelihood ratio tests (Zuur et al. 2009; Bolker et al. 2009). The R package nlme was used (Pinheiro et al. 2015). The estimates were taken from a model with arrival date not square-root transformed for a better understanding of the effect, and the t- and p-values from a model where arrival date was square-root transformed to obtain a better fit to a normal distribution. To predict microevolutionary change over time we used the breeder’s equation (Lush 1937). For
the breeder’s equation (R=h2*S, where the evolutionary response per generation, R, equals the
heritability, h2, times the selection differential S), we fitted the selection differential based on LRS (recalculated to day units) and genetic variance standardized by the phenotypic variance (i.e. heritability, h2) because arrival date is on an interval scale (Houle et al. 2011; Hansen et al. 2011). To visualize the expected microevolutionary change based on the breeder’s equation in relation to the phenotypic change and shift in spring warming onset, we calculated the cumulative expected change per year (dividing the per generation change with the generation time) with 1985’s mean bird arrival date as the starting phenotype. We note that there are several potential problems with the prediction of microevolution using the breeder’s equation, including the problem to infer the causality of the correlation between trait and fitness in unmanipulated wild populations (Morrissey et al. 2010). However, in many cases, this is the best prediction that can be done.
To complement the prediction from the breeder’s equation, we compared it with the prediction
from the Robertson-Price identity (Morrissey et al. 2010). The selection differential can sometimes be driven by environmental factors, such as condition, that affect both the phenotype and fitness and create a non-genetic correlation (cf. Price et al. 1988b; Merilä et al. 2001). By using the Robertson-Price Identity this potential problem can be avoided and selection acting directly on the genetic component of
This article is protected by copyright. All rights reserved.
Accepted Article
a trait is estimated (Morrissey et al. 2010). To distinguish if selection was acting on the genetic or the environmental component of arrival date, we fitted a bivariate ‘animal model’ with relative fitness and
individual mean arrival date as response variables. Here, we used the same data set as for the selection estimates (corrected (age, year) individual mean arrival date (here not transformed to mean 0 and SD 1) and relative lifetime recruit success calculated as above). The fitted model included sex as a fixed factor and additive genetic effect as a random factor. We used the software AsReml v.3.0 (Gilmour et al. 2009)
and significance of correlations was obtained by using log-likelihood ratio test between a model with the covariance varying freely and a model where the covariance was constrained to zero (Wilson et al. 2010).
Fine-grained population responses to environmental changes can be used to infer phenotypic plasticity as a likely explanation for phenotypic changes over time. If the population mean strongly correlates with an environmental trait on a year-to-year basis, phenotypic plasticity becomes a likely explanation (Merilä & Hendry 2014). To investigate how population mean arrival date correlates with yearly fluctuations in the local environment at the breeding site, we calculated the yearly onset of spring warming in Kvismaren. To obtain a standardized timing of the start of the spring, i.e. onset of spring warming, at the breeding site each year, we used a similar approach to that of (Saino et al. 2011) based on a calculation of accumulating winter and spring temperatures. Each year, when the cumulative sum of the daily mean temperatures reached above a set level (see below), the date when this occurred was used as the date for the onset of spring warming. Daily mean temperatures for the years 1980-2004 were downloaded from the Swedish Meteorological and Hydrological Institute (www.SMHI.se; assessed September 2013). We used data from three weather stations in the area around our study site (Örebro: 59°28’N, 15°16’E; Karlstad Flygplats: 59°44’N, 13°34’E; Jönköpings Flygplats: 57°75’N, 14°07’E) and
This article is protected by copyright. All rights reserved.
Accepted Article
calculated the temperature means for each day based on these. Then we calculated the accumulation of ‘degree-days’ (DD) with a threshold 0°C for the instantaneous temperatures (Saino et al. 2011) and the spring ‘mean arrival date’ (MAD) based on raw data of the yearly averages of the years 1985-1987. This MAD was then used to calculate DD for the years 1980-1984 and the average DD of these years was then used as a baseline (DDbase) for the spring-onset defining temperatures. For each of the study years 1985-2004, we cumulatively summed the temperatures (above 0°C) from 1st of January until DDbase was reached. The date when the temperature passed the DDbase threshold was used as the date of the onset of spring warming for that particular year.
Results Mean uncorrected arrival date for males was 19th of May (SD = 14 days, N=253) and for females 30th of May (SD = 12 days, N=295). Mean arrival date of each year’s first male was 3rd of May (SD = 3 days) and
for the first female it was 14th of May (SD = 5 days).
Selection on arrival date We found negative directional selection on arrival date for the sexes combined, as well as for males and females separately, both when using lifetime fledgling success (LFS) and lifetime recruiting offspring success (LRS) as fitness estimates (estimates of standardized selection differentials ranging from -0.226
± 0.058 to -0.368 ± 0.099; Table 1). This means that in both sexes, individuals that arrived earlier in
spring produced relatively more fledglings and recruiting offspring over their lives than the population average. The estimates and standard errors from these individual mean models (sexes combined; LFS: 0.267 ± 0.052; LRS: -0.341 ± 0.071; see Table 1) were very similar to estimates from models using
This article is protected by copyright. All rights reserved.
Accepted Article
repeated measures (sexes combined; Annual FS: -0.280 ± 0.033; Annual RS: -0.332 ± 0.065). For lifespan, the directional selection estimates on arrival date were lower (ranging from -0.021 to -0.051) and non-significant (Table 1). When analysing quadratic selection on arrival date with respect to lifespan, there was a
significant stabilizing selection in females and for the sexes combined. In males, the estimate of stabilizing selection was of the same magnitude as in females; however it was not significant (Table 1). One problem with using individual means to estimate quadratic selection is that individuals that live longer will be repeatedly measured and therefore their mean values are expected to vary less compared to short lived individuals that only occur for one or a few years in the population. This unequal variance between short- and long-lived birds could potentially generate a pattern between arrival date and lifespan that resembles stabilizing selection (individuals with high lifespan have values with low variance, and individuals with short lifespan have values with high variance, around the mean standardized arrival date, and when a curve is fitted to such data it will include a negative quadratic component). To avoid this potentially spurious effect, we reran our arrival date–lifespan analysis including only (year and age standardized) arrival date from one random year for each individual. This analysis showed that in all three cases (males, females and sexes combined) the quadratic estimates of lifespan selection were lower in magnitude compared with the analysis based on means, and also not statistically significant (quadratic selection differential estimates (doubled) with 95% CI [lower, upper]; males: 0.024 [-0.072, 0.114]; females: -0.038 [-0.114, 0.42]; total: -0.010 [-0.070, 0.050]). This suggests that the significant estimate of quadratic selection acting through lifespan might be inflated and therefore should be treated with caution. Cubic splines that visualise the selection acting on arrival date (Figure 1a-c) largely confirmed
the patterns found in the analyses of selection differentials (Table 1). For LFS (Figure 1a) and LRS (Figure
This article is protected by copyright. All rights reserved.
Accepted Article
1b), the curve had a negative slope with a peak left of the population mean, which indicates that there is pronounced negative directional selection acting on arrival date through these two fitness estimates. Regarding lifespan (Figure 1c), the curve was more hump-shaped and the optimum is closer to the population mean, which is in line with the analyses of selection differentials where we found weak nonsignificant directional selection and some evidence in support of stabilizing selection (Table 1, but see the caveats above).
Repeatability and heritability of arrival date Repeatability of arrival date was moderate and significant (0.366 ± 0.040 SE, P < 0.001), meaning that 36.6% of the phenotypic variance is explained by individuals having consistent arrival date between years. Furthermore, we found a significant heritability of arrival date of 16.4% (h2 = 0.164 ± 0.065 SE, P
