Avian seasonal timing is a life-history trait with important fitness consequences and which is currently under directional selection due to climate change. To predict micro-evolution in this trait, it is crucial to properly estimate its heritability. Heritabilities are often estimated from pedigreed wild populations. As these are observational data, it leaves the possibility that the resemblance between related individuals is not due to shared genes but to ontogenetic effects; when the environment for the offspring provided by early laying pairs differs from that by late pairs and the laying dates of these offspring when they reproduce themselves is affected by this environment, this may lead to inflated heritability estimates. Using simulation studies, we first tested whether and how much such an early environmental effect can inflate heritability estimates from animal models, and we showed that pedigree structure determines by how much early environmental effects inflate heritability estimates. We then used data from a wild population of great tits (Parus major) to compare laying dates of females born early in the season in first broods and from sisters born much later, in second broods. These birds are raised under very different environmental conditions but have the same genetic background. The laying dates of first and second brood offspring do not differ when they reproduce themselves, clearly showing that ontogenetic effects are very small and hence, family resemblance in timing is due to genes. This finding is essential for the interpretation of the heritabilities reported from wild populations and for predicting micro-evolution in response to climate change.
In birds, seasonal timing of breeding has strong consequences for reproductive success (e.g. Daan et al., 1990; Brinkhof et al., 1993; Verhulst et al., 1995). An important factor determining reproductive success is food availability for chicks (van Noordwijk et al., 1995; Visser et al., 2006). Climate change has disrupted the synchrony between timing of breeding and food availability in a larger number of studies (reviewed in Visser & Both, 2005), and this phenological mismatch has led to selection for earlier breeding (Visser et al., 1998). To be able to assess whether populations will be able to respond to this selection – and how fast – we need reliable estimates of genetic variation in seasonal timing.
Avian breeding time has been found to be heritable (e.g. van Noordwijk et al., 1981; Sheldon et al., 2003; Gienapp et al., 2006). These estimates are however based on phenotypic resemblance among relatives, and they could hence be inflated by environmentally caused covariance in phenotypes among relatives. For example, spatial autocorrelation of environmental conditions could lead to inflated phenotypic similarity among relatives since dispersal distances are limited and relatives are hence likely to breed in similar environments (van der Jeugd & McCleery, 2002). Other possible effects causing an environmental covariance among relatives could be ontogenetic effects. We know that physiological condition affects several aspects of breeding (e.g. Horak et al., 1998; Dubiec et al., 2005; Ardia et al., 2006). The decreasing food supply during the season could affect offspring condition directly or via maternal effects because late-breeding females are already in poorer condition (Verboven et al., 2003; Dubiec et al., 2005). Offspring born late in a season often have a lighter fledging weight (van Noordwijk et al., 1988; Gebhardt-Henrich, 1990; Perrins & McCleery, 2001; Visser et al., 2006). This initial seasonal effect on condition might further be amplified by the fact that these individuals are more likely to obtain a low position in the hierarchy of winter flocks which could easily translate to a poor physiological condition in spring (Visser & Verboven, 1999; Perrins & McCleery, 2001). Carry-over effects from one stage in the annual cycle to the next have also been demonstrated. For example, the speed of moult and the age at the onset of autumn migration are very different for birds that hatched early and those that hatched late (e.g. Helm & Gwinner, 1999; Coppack & Pulido, 2004). Consequently, we might expect that subtle effects during ontogeny could lead to late born offspring being likely to breed late themselves and hence an environmentally caused resemblance in timing of breeding among relatives.
Early environmental effects can also confound heritabilities in other traits. For example, if preferences for habitats or mating partners were ‘learned’ during ontogeny, offspring behaviour would resemble the behaviour of their parents which would obviously inflate any estimate of the heritability of these traits (Stamps et al., 2009).
Testing whether ontogenetic effects are responsible for a similarity in timing of breeding among relatives is however problematic using descriptive data from natural populations since ontogenetic and genetic effect are here confounded. Whereas ontogenetic effects of other traits, as for example habitat preference of clutch size can easily be studied by experimental manipulations, it is not feasible to substantially manipulate the timing (Verhulst & Nilsson, 2008).
Environmental variance that could inflate estimates of genetic (co)variances is a general problem in studies of natural populations (Lynch & Walsh, 1998). Using information from more distant relatives in an ‘animal model’ can reduce this bias (Kruuk & Hadfield, 2007), especially when spatial environmental covariance is a potential problem because more distant relatives are more likely to breed in a different environment (van der Jeugd & McCleery, 2002). However, if the hatching date of an individual affects its own egg-laying date later in life, this possible confounding effect will be exactly the same for all individuals because all individuals hatched at the same time will experience the same seasonal environment. For example, the early or late hatched individuals will experience the same short or long day lengths, respectively, which can affect the individual's circadian clock later in life (Ciarleglio et al., 2011). How well an ‘animal model’ will be able to separate this confounding early environment effect from an additive genetic effect depends on the pedigree structure, that is, the amount of information available on relatives, but comprehensive studies on this are still lacking.
We here used a two-step approach to tackle this problem. First, we used simulation studies to assess how powerful ‘animal models’ are to separate such early environment effects from the additive genetic effect. Second, we use a data set on a wild bird population to disentangle early environmental or ontogenetic and (additive) genetic effects by comparing timing of breeding of female great tits that were born in first and second broods of the same female. In some populations great tits regularly produce first and second broods in one breeding season (Verboven et al., 2001), that is, raising another brood after already having successfully raised one. Hatching dates of offspring from first and second brood differ by about 6 weeks. Comparing timing of breeding of daughters from first and second broods offers hence a unique opportunity to test for ontogenetic effects since these individuals differ substantially in the timing when they were born and raised but not in their genetic background (van Noordwijk, 2006).
Materials and methods
To test whether an ‘animal model’ would be able to reliably disentangle early environmental effects from the additive genetic effect, we ran two simulation models for egg-laying dates. In the first model, the ‘genetic model’, egg-laying dates were determined by an additive genetic effect. In the second model, the ‘early environment model’, egg-laying dates depended solely on the individual's own hatching date, that is, its mothers’ egg-laying dates. The variance explained by this early environment effect was identical to the additive genetic variance and hence the ‘heritability’ of egg-laying date should vary from zero, if the ‘animal model’ could separate the two effects perfectly, and the ‘true’ heritability, if the ‘animal model’ would completely fail to separate the two effects.
To test the effect of pedigree structure on the power of the ‘animal model’, we simulated data for two quite different pedigrees from our long-term study populations: Vlieland and the Hoge Veluwe. While Vlieland is an island population with consequently low immigration rates, the Hoge Veluwe study area is part of a large continuous woodland with higher immigration rates. This has clear effects on pedigree structure: While the absolute pedigree sizes are comparable (Vlieland: 6482, Hoge Veluwe: 6292), the proportion of ‘founders’, that is, individuals with unknown parents, was much lower on Vlieland (0.30) than in the Hoge Veluwe (0.63). The maximum pedigree depth differed also considerably with 30 generations for Vlieland but only 16 generations for the Hoge Veluwe.
Data for the ‘genetic model’ were simulated by simulating breeding values for both females and males with an additive genetic variance (VA) of 4.3 through the two pedigrees using the ‘rbv’ function from the MCMCglmm package (Hadfield, 2010). Phenotypes were then simulated by adding an environmental component, a random number drawn from a normal distribution with mean zero and variance 10.8 (VRes). This values give a heritability of 0.285, the mean of the two subpopulations on Vlieland. Since males do not affect their females’ egg-laying dates (Caro et al., 2009), males were not assigned phenotypes. However, in this model, we assumed that males would carry genes for egg-laying dates, and they were hence assigned breeding values.
Data for the ‘early environment model’ were simulated by assigning all females in the base population, that is, without known ancestors, a random egg-laying date drawn from a normal distribution with mean zero and variance equal to the total phenotypic variance, that is, VA plus VRes. Egg-laying dates were then simulated through the pedigree so that the mother–daughter resemblance was equal to h2/2. Simulations for both models and both pedigrees were run 500 times. The R code used for these simulation models is available as Appendix S1.
The output from the models was analysed with an ‘animal model’ including only the additive genetic effect as every individual bred only once, and we did not model variation in egg-laying dates among years or age classes. The ‘animal models’ were run with MCMCglmm (Hadfield, 2010) in R 2.14.1 (R Development Core Team, 2011) using the default flat priors.
Study population and field work
The great tit (Parus major) population on the island of Vlieland (53°17′N, 5°3′E) in the Dutch Wadden Sea has been monitored continuously since 1955. Nest boxes are provided in excess in all suitable nesting habitat. The population breeds in two distinct subpopulations that differ little in breeding time and substantially in clutch size (Postma & van Noordwijk, 2005; Postma et al., 2007). During the breeding season, nest boxes were checked weekly and all breeding attempts, first clutches, replacement clutches (following a failed first brood) and second clutches (following a successful first brood), recorded. The laying date of the first egg of a clutch (hereafter egg-laying date) was calculated from the number of eggs found during these checks, assuming that one egg is laid per day. Breeding time has been found to be heritable in the two subpopulations using an animal model (h2 = 0.23 ± 0.04 and 0.34 ± 0.08 for the two subpopulations Postma, 2005). Hatching dates were calculated from age estimates of chicks at first visit after hatching. Hatching dates of first and second broods differed by 43.1 days (SD = 5.4, n = 823). Adults were caught during chick feeding and identified by their aluminium and colour rings or ringed if not previously caught. All nestlings were ringed with aluminium rings before fledging.
Extra-pair paternity in great tits on Vlieland is low (Verboven & Mateman, 1997) and pairs generally (95.5%, n = 823) stay together between first and second broods. Between years 46.3% (n = 1510) of the females keep the same partner. Consequently, most daughters from first and second broods within years are full-siblings.
To test whether time of rearing of a female had an effect on her egg-laying dates later in life, we used two approaches. First, we tested whether the heritabilities obtained from mother–daughter regression differed for daughters from first broods and daughters from second broods using standard linear models. If hatching date affects the recruits’ own egg-laying dates, we would expect that egg-laying dates of daughters from first broods to more closely resemble their mothers egg-laying dates, that is, have a higher heritability. We had data from 797 females with female recruits from first broods and 124 females with female recruits from second broods. If a female recruited more than one daughter, the daughters’ egg-laying dates were averaged. Second, in a more direct and stronger test, we compared egg-laying dates in the first year of breeding of females born in first and second broods of the same mother. From 80 females breeding between 1959 and 2010, 152 daughters from a first and 95 from a second brood recruited and have a known laying date of their first clutch in their first year and were included into the analyses. Data were analysed in a mixed-model with mother as random effect. All egg-laying dates were corrected (centred) for year-to-year variation in breeding time by subtracting the respective annual mean egg-laying date of all first broods. Significance of fixed effects in mixed models was calculated using Kenward–Roger approximation in R 2.14.1 (R Development Core Team, 2011).
The heritabilities of egg-laying dates calculated from mother–daughter regressions were very similar for both models and pedigrees (Fig. 1). The results from the ‘animal model’ differed however strikingly depending on pedigree structure: for the ‘deeper’ Vlieland-pedigree the ‘heritability’ from the ‘early environment model’ was much smaller (0.079 ± 0.024, mean ± SD from 500 simulation runs) than the heritability from the ‘genetic model’ (0.28 ± 0.04, mean ± SD) (Fig. 2). For the Hoge Veluwe-pedigree the ‘heritability’ from the ‘early environment model’ was however much more similar to the heritability from the ‘genetic model’ (0.17 ± 0.09 vs. 0.27 ± 0.07, mean ± SD) (Fig. 2). We refrained from formally testing these differences because sample size can be arbitrarily set which makes statistical tests meaningless in theoretical models.
Egg-laying dates of offspring from first and second broods
The heritability, estimated from mother–daughter regression, was 0.27 (F1,919 = 20.5, P < 0.001). There was no statistical significant difference in heritability for daughters from second broods (h2 = 0.19) and daughters from first broods (h2 = 0.28); the interaction between the mother's egg-laying date and type of brood (first vs. second) (Fig. 3) was not significant (F1,917 = 0.28, P = 0.60). There was no evidence either that type of brood would affect egg-laying dates over and above the effect of the mother's egg-laying date (b =0.22 ± 0.44, F1,918 = 0.26, P = 0.61). The variation in egg-laying dates of daughters from second broods was larger (29.7) than that in egg-laying dates of daughters from first broods (19.6).
Daughters from second broods laid on average 0.69 ± 0.70 (se) days later than their (half-)sisters who were born in a first brood (Fig. 4). This difference was however not significant (F1,206.0 = 0.96, P = 0.33). Given a sample size of 80 females having recruited daughters from both first and second clutches and the observed se of the difference we could have detected ontogenetic effects of about 1.5 days based on the rule of thumb for paired t-tests that a difference of 1.96*se is significant with P = 0.05.
The results from our simulation models clearly show that an early environmental effect of the individuals’ own hatching date would be indistinguishable with data from only two generations (Fig. 1). As expected the ‘animal model’ performs better in separating such an early environmental or ontogenetic effect from a real additive genetic effect but this depends quite strongly on the amount of information that is coming from relatives, that is, the pedigree structure (Fig. 2). For the Vlieland-pedigree the ‘heritability’ for the ‘early environment model’ was less than a third of the heritability estimated for the ‘genetic model’, but still larger than zero. For the shallower Hoge Veluwe-pedigree, the ‘heritability’ for the ‘early environment model’ was 62% of the heritability estimated for the ‘genetic model’ and well in the order of magnitude for heritabilities for egg-laying date and similar such life-history traits (e.g. van Noordwijk et al., 1981; Sheldon et al., 2003; McCleery et al., 2004; Gienapp et al., 2006).
Egg-laying dates are under female control (Caro et al., 2009), but it is currently unknown whether this is a sex-linked trait, that is, only females would carry genes for egg-laying date. In our simulation model, we assumed this not to be the case, that is, egg-laying dates were inherited through males as well. Inheritance through the male line obviously gives additional important information to disentangle additive genetic effects from maternal effects. It is hence be somewhat surprising that the ‘animal model’ did still so comparably poorly in disentangling these effects for the Hoge Veluwe-pedigree. If we had assumed that egg-laying date were a sex-linked trait in our simulation, this bias would have been even stronger (in both populations) and may have rendered the ‘animal model’ unable to separate the ‘early environmental’ from the genetic effect for the Hoge Veluwe-pedigree.
The pedigree for the great tit population on Vlieland is complex due to the low dispersal possibilities from and into this population. The shallower and less complex pedigree structure in the Hoge Veluwe population probably resembles better the situation in most wild pedigreed study populations, although there is very little information on pedigree structure published. Consequently, we should expect that early environmental or ontogenetic effects, but also maternal effects, can inflate quantitative genetic parameters. The possible bias in quantitative genetic estimates due to such factors has been generally acknowledged (e.g. Lynch & Walsh, 1998; Kruuk, 2004). Effects of spatially or temporally correlated environmental conditions have also been tested experimentally and empirically (van der Jeugd & McCleery, 2002; Kruuk & Hadfield, 2007). We are however still lacking a general quantitative assessment how unidentified maternal, environmental or other effects could bias quantitative genetic estimates obtained from natural populations. Further simulation studies, as our analysis here, assessing a variety of effects that can possibly bias such estimates and a range of realistic pedigree structures are hence necessary to address this problem more comprehensively.
Our results from a wild bird population that egg-laying dates of daughters born to the same mother but in first and second broods did not differ substantially indicates that (early) ontogenetic effects do not have a measurable effect on subsequent timing of breeding. Daughters from first broods hatched on average 6 weeks before their sisters from second broods and hence experience a very different environment which could affect their subsequent breeding performance. For example, the declining abundance of caterpillars, an important prey species for nestlings (Perrins, 1991; Naef-Daenzer et al., 2000), during the breeding season leads to a reduced fledging weight of late born individuals (Gebhardt-Henrich, 1990; Visser et al., 2006). This or other, possibly more subtle, ontogenetic effects (e.g. Ciarleglio et al., 2011) do however not seem to affect egg-laying dates in the following year. Consequently, early ontogenetic effects caused by environmental differences within the range of first broods are not a likely cause of inflated heritability estimates of breeding time.
Our finding is important for understanding, and predicting, micro-evolution because inflated heritabilities will lead to over-estimated evolutionary potential, that is, predicted possible response to selection. This is especially relevant in a context of persistent environmental change, such as global climate change, that will lead to sustained, directional selection which ultimately requires evolutionary adaptation to avoid increased extinction risk of populations or even species (Lynch & Lande, 1993; Bürger & Lynch, 1995). Precisely assessing the potential for ‘evolutionary rescue’ is hence important to identify populations or species possibly at elevated risk of extinction (Bell & Gonzalez, 2009; Willi & Hoffmann, 2009; Gienapp et al., 2013). As our simulation model showed, early environmental effects can inflate heritability estimates, even in analyses based on the animal model depending on pedigree structure. However, by comparing timing of breeding between sisters from first and second clutches, we could elegantly break the correlation between genes for breeding early and being raised early in the season and thereby ascertain that ontogenetic effects do not inflate heritability estimates of breeding time in great tits.
Very many people have contributed to the collection of the data, we want to explicitly thank Jan Visser, Elsje de Ruiter and Louis Vernooij for field work and database maintenance as well as Staatsbosbeheer for permission to work in forests on Vlieland. An anonymous referee inspired us to simulate how well an animal model can distinguish genetic from early environment effects. MEV is supported by a NWO-VICI grant.